From 41320a0554716aaf7cec6172da98e002c48344c5 Mon Sep 17 00:00:00 2001 From: Nick Mills-Barrett Date: Thu, 4 Aug 2022 15:49:55 +0100 Subject: Optimise async get event lookups (#13435) Still maintains local in memory lookup optimisation, but does any external lookup as part of the deferred that prevents duplicate lookups for the same event at once. This makes the assumption that fetching from an external cache is a non-zero load operation. --- synapse/util/caches/lrucache.py | 17 +++++++++++++++++ 1 file changed, 17 insertions(+) (limited to 'synapse/util') diff --git a/synapse/util/caches/lrucache.py b/synapse/util/caches/lrucache.py index b3bdedb04c..aa93109d13 100644 --- a/synapse/util/caches/lrucache.py +++ b/synapse/util/caches/lrucache.py @@ -834,9 +834,26 @@ class AsyncLruCache(Generic[KT, VT]): ) -> Optional[VT]: return self._lru_cache.get(key, update_metrics=update_metrics) + async def get_external( + self, + key: KT, + default: Optional[T] = None, + update_metrics: bool = True, + ) -> Optional[VT]: + # This method should fetch from any configured external cache, in this case noop. + return None + + def get_local( + self, key: KT, default: Optional[T] = None, update_metrics: bool = True + ) -> Optional[VT]: + return self._lru_cache.get(key, update_metrics=update_metrics) + async def set(self, key: KT, value: VT) -> None: self._lru_cache.set(key, value) + def set_local(self, key: KT, value: VT) -> None: + self._lru_cache.set(key, value) + async def invalidate(self, key: KT) -> None: # This method should invalidate any external cache and then invalidate the LruCache. return self._lru_cache.invalidate(key) -- cgit 1.5.1 From 344a2f767c636259412f7fc2914c1554a5c4dc1d Mon Sep 17 00:00:00 2001 From: Eric Eastwood Date: Mon, 15 Aug 2022 13:41:23 -0500 Subject: Instrument `FederationStateIdsServlet` - `/state_ids` (#13499) Instrument FederationStateIdsServlet - `/state_ids` so it's easier to follow what's going on in Jaeger when viewing a trace. --- changelog.d/13499.misc | 1 + synapse/federation/federation_server.py | 11 ++++++++++- synapse/handlers/federation.py | 4 +++- synapse/storage/databases/main/event_federation.py | 3 +++ synapse/util/ratelimitutils.py | 4 ++++ 5 files changed, 21 insertions(+), 2 deletions(-) create mode 100644 changelog.d/13499.misc (limited to 'synapse/util') diff --git a/changelog.d/13499.misc b/changelog.d/13499.misc new file mode 100644 index 0000000000..99dbcebec8 --- /dev/null +++ b/changelog.d/13499.misc @@ -0,0 +1 @@ +Instrument `FederationStateIdsServlet` (`/state_ids`) for understandable traces in Jaeger. diff --git a/synapse/federation/federation_server.py b/synapse/federation/federation_server.py index db4b83a505..75fbc6073d 100644 --- a/synapse/federation/federation_server.py +++ b/synapse/federation/federation_server.py @@ -61,7 +61,12 @@ from synapse.logging.context import ( nested_logging_context, run_in_background, ) -from synapse.logging.opentracing import log_kv, start_active_span_from_edu, trace +from synapse.logging.opentracing import ( + log_kv, + start_active_span_from_edu, + tag_args, + trace, +) from synapse.metrics.background_process_metrics import wrap_as_background_process from synapse.replication.http.federation import ( ReplicationFederationSendEduRestServlet, @@ -547,6 +552,8 @@ class FederationServer(FederationBase): return 200, resp + @trace + @tag_args async def on_state_ids_request( self, origin: str, room_id: str, event_id: str ) -> Tuple[int, JsonDict]: @@ -569,6 +576,8 @@ class FederationServer(FederationBase): return 200, resp + @trace + @tag_args async def _on_state_ids_request_compute( self, room_id: str, event_id: str ) -> JsonDict: diff --git a/synapse/handlers/federation.py b/synapse/handlers/federation.py index 5042236742..6f5ab86ac4 100644 --- a/synapse/handlers/federation.py +++ b/synapse/handlers/federation.py @@ -59,7 +59,7 @@ from synapse.events.validator import EventValidator from synapse.federation.federation_client import InvalidResponseError from synapse.http.servlet import assert_params_in_dict from synapse.logging.context import nested_logging_context -from synapse.logging.opentracing import trace +from synapse.logging.opentracing import tag_args, trace from synapse.metrics.background_process_metrics import run_as_background_process from synapse.module_api import NOT_SPAM from synapse.replication.http.federation import ( @@ -1081,6 +1081,8 @@ class FederationHandler: return event + @trace + @tag_args async def get_state_ids_for_pdu(self, room_id: str, event_id: str) -> List[str]: """Returns the state at the event. i.e. not including said event.""" event = await self.store.get_event(event_id, check_room_id=room_id) diff --git a/synapse/storage/databases/main/event_federation.py b/synapse/storage/databases/main/event_federation.py index eec55b6478..0bc8401f2b 100644 --- a/synapse/storage/databases/main/event_federation.py +++ b/synapse/storage/databases/main/event_federation.py @@ -33,6 +33,7 @@ from synapse.api.constants import MAX_DEPTH, EventTypes from synapse.api.errors import StoreError from synapse.api.room_versions import EventFormatVersions, RoomVersion from synapse.events import EventBase, make_event_from_dict +from synapse.logging.opentracing import tag_args, trace from synapse.metrics.background_process_metrics import wrap_as_background_process from synapse.storage._base import SQLBaseStore, db_to_json, make_in_list_sql_clause from synapse.storage.database import ( @@ -126,6 +127,8 @@ class EventFederationWorkerStore(SignatureWorkerStore, EventsWorkerStore, SQLBas ) return await self.get_events_as_list(event_ids) + @trace + @tag_args async def get_auth_chain_ids( self, room_id: str, diff --git a/synapse/util/ratelimitutils.py b/synapse/util/ratelimitutils.py index 6394cc39ac..e1beaec5a3 100644 --- a/synapse/util/ratelimitutils.py +++ b/synapse/util/ratelimitutils.py @@ -27,6 +27,7 @@ from synapse.logging.context import ( make_deferred_yieldable, run_in_background, ) +from synapse.logging.opentracing import start_active_span from synapse.util import Clock if typing.TYPE_CHECKING: @@ -176,8 +177,11 @@ class _PerHostRatelimiter: # Ensure that we've properly cleaned up. self.sleeping_requests.discard(request_id) self.ready_request_queue.pop(request_id, None) + wait_span_scope.__exit__(None, None, None) return r + wait_span_scope = start_active_span("ratelimit wait") + wait_span_scope.__enter__() ret_defer.addCallbacks(on_start, on_err) ret_defer.addBoth(on_both) return make_deferred_yieldable(ret_defer) -- cgit 1.5.1 From c6ee9c0ee40803a9e3673c2833e5a40032e86f5a Mon Sep 17 00:00:00 2001 From: Eric Eastwood Date: Wed, 17 Aug 2022 04:38:05 -0500 Subject: Add metrics to track rate limiter queue timing (#13544) --- changelog.d/13544.misc | 1 + synapse/util/ratelimitutils.py | 30 ++++++++++++++++++++++++++++++ 2 files changed, 31 insertions(+) create mode 100644 changelog.d/13544.misc (limited to 'synapse/util') diff --git a/changelog.d/13544.misc b/changelog.d/13544.misc new file mode 100644 index 0000000000..d84ba3f076 --- /dev/null +++ b/changelog.d/13544.misc @@ -0,0 +1 @@ +Add metrics to track rate limiter queue timing (`synapse_rate_limit_queue_wait_time_seconds`). diff --git a/synapse/util/ratelimitutils.py b/synapse/util/ratelimitutils.py index e1beaec5a3..e48324d926 100644 --- a/synapse/util/ratelimitutils.py +++ b/synapse/util/ratelimitutils.py @@ -28,6 +28,7 @@ from synapse.logging.context import ( run_in_background, ) from synapse.logging.opentracing import start_active_span +from synapse.metrics import Histogram from synapse.util import Clock if typing.TYPE_CHECKING: @@ -36,6 +37,29 @@ if typing.TYPE_CHECKING: logger = logging.getLogger(__name__) +queue_wait_timer = Histogram( + "synapse_rate_limit_queue_wait_time_seconds", + "sec", + [], + buckets=( + 0.005, + 0.01, + 0.025, + 0.05, + 0.1, + 0.25, + 0.5, + 0.75, + 1.0, + 2.5, + 5.0, + 10.0, + 20.0, + "+Inf", + ), +) + + class FederationRateLimiter: def __init__(self, clock: Clock, config: FederationRatelimitSettings): def new_limiter() -> "_PerHostRatelimiter": @@ -178,10 +202,16 @@ class _PerHostRatelimiter: self.sleeping_requests.discard(request_id) self.ready_request_queue.pop(request_id, None) wait_span_scope.__exit__(None, None, None) + wait_timer_cm.__exit__(None, None, None) return r + # Tracing wait_span_scope = start_active_span("ratelimit wait") wait_span_scope.__enter__() + # Metrics + wait_timer_cm = queue_wait_timer.time() + wait_timer_cm.__enter__() + ret_defer.addCallbacks(on_start, on_err) ret_defer.addBoth(on_both) return make_deferred_yieldable(ret_defer) -- cgit 1.5.1 From 49d04e43dfa0551f52c1f1872b6f311efa756ca8 Mon Sep 17 00:00:00 2001 From: Eric Eastwood Date: Wed, 17 Aug 2022 16:10:07 -0500 Subject: Add metrics to track how the rate limiter is affecting requests (sleep/reject) (#13534) Related to https://github.com/matrix-org/synapse/pull/13499 Part of https://github.com/matrix-org/synapse/issues/13356 --- changelog.d/13534.misc | 1 + synapse/util/ratelimitutils.py | 37 +++++++++++++++++++++++++++++-------- 2 files changed, 30 insertions(+), 8 deletions(-) create mode 100644 changelog.d/13534.misc (limited to 'synapse/util') diff --git a/changelog.d/13534.misc b/changelog.d/13534.misc new file mode 100644 index 0000000000..b488bf74c3 --- /dev/null +++ b/changelog.d/13534.misc @@ -0,0 +1 @@ +Add metrics to track how the rate limiter is affecting requests (sleep/reject). diff --git a/synapse/util/ratelimitutils.py b/synapse/util/ratelimitutils.py index e48324d926..434b02b97b 100644 --- a/synapse/util/ratelimitutils.py +++ b/synapse/util/ratelimitutils.py @@ -18,6 +18,8 @@ import logging import typing from typing import Any, DefaultDict, Iterator, List, Set +from prometheus_client.core import Counter + from twisted.internet import defer from synapse.api.errors import LimitExceededError @@ -37,6 +39,9 @@ if typing.TYPE_CHECKING: logger = logging.getLogger(__name__) +# Track how much the ratelimiter is affecting requests +rate_limit_sleep_counter = Counter("synapse_rate_limit_sleep", "") +rate_limit_reject_counter = Counter("synapse_rate_limit_reject", "") queue_wait_timer = Histogram( "synapse_rate_limit_queue_wait_time_seconds", "sec", @@ -84,7 +89,7 @@ class FederationRateLimiter: Returns: context manager which returns a deferred. """ - return self.ratelimiters[host].ratelimit() + return self.ratelimiters[host].ratelimit(host) class _PerHostRatelimiter: @@ -119,12 +124,14 @@ class _PerHostRatelimiter: self.request_times: List[int] = [] @contextlib.contextmanager - def ratelimit(self) -> "Iterator[defer.Deferred[None]]": + def ratelimit(self, host: str) -> "Iterator[defer.Deferred[None]]": # `contextlib.contextmanager` takes a generator and turns it into a # context manager. The generator should only yield once with a value # to be returned by manager. # Exceptions will be reraised at the yield. + self.host = host + request_id = object() ret = self._on_enter(request_id) try: @@ -144,6 +151,8 @@ class _PerHostRatelimiter: # sleeping or in the ready queue). queue_size = len(self.ready_request_queue) + len(self.sleeping_requests) if queue_size > self.reject_limit: + logger.debug("Ratelimiter(%s): rejecting request", self.host) + rate_limit_reject_counter.inc() raise LimitExceededError( retry_after_ms=int(self.window_size / self.sleep_limit) ) @@ -155,7 +164,8 @@ class _PerHostRatelimiter: queue_defer: defer.Deferred[None] = defer.Deferred() self.ready_request_queue[request_id] = queue_defer logger.info( - "Ratelimiter: queueing request (queue now %i items)", + "Ratelimiter(%s): queueing request (queue now %i items)", + self.host, len(self.ready_request_queue), ) @@ -164,19 +174,28 @@ class _PerHostRatelimiter: return defer.succeed(None) logger.debug( - "Ratelimit [%s]: len(self.request_times)=%d", + "Ratelimit(%s) [%s]: len(self.request_times)=%d", + self.host, id(request_id), len(self.request_times), ) if len(self.request_times) > self.sleep_limit: - logger.debug("Ratelimiter: sleeping request for %f sec", self.sleep_sec) + logger.debug( + "Ratelimiter(%s) [%s]: sleeping request for %f sec", + self.host, + id(request_id), + self.sleep_sec, + ) + rate_limit_sleep_counter.inc() ret_defer = run_in_background(self.clock.sleep, self.sleep_sec) self.sleeping_requests.add(request_id) def on_wait_finished(_: Any) -> "defer.Deferred[None]": - logger.debug("Ratelimit [%s]: Finished sleeping", id(request_id)) + logger.debug( + "Ratelimit(%s) [%s]: Finished sleeping", self.host, id(request_id) + ) self.sleeping_requests.discard(request_id) queue_defer = queue_request() return queue_defer @@ -186,7 +205,9 @@ class _PerHostRatelimiter: ret_defer = queue_request() def on_start(r: object) -> object: - logger.debug("Ratelimit [%s]: Processing req", id(request_id)) + logger.debug( + "Ratelimit(%s) [%s]: Processing req", self.host, id(request_id) + ) self.current_processing.add(request_id) return r @@ -217,7 +238,7 @@ class _PerHostRatelimiter: return make_deferred_yieldable(ret_defer) def _on_exit(self, request_id: object) -> None: - logger.debug("Ratelimit [%s]: Processed req", id(request_id)) + logger.debug("Ratelimit(%s) [%s]: Processed req", self.host, id(request_id)) self.current_processing.discard(request_id) try: # start processing the next item on the queue. -- cgit 1.5.1 From d64653d062a7fc27782e70c1ca581e85b7730e72 Mon Sep 17 00:00:00 2001 From: Eric Eastwood Date: Thu, 18 Aug 2022 10:05:07 -0500 Subject: Track number of hosts affected by the rate limiter (#13541) Track number of hosts affected by the rate limiter so we can differentiate one really noisy homeserver from a general ratelimit tuning problem across the federation. Follow-up to https://github.com/matrix-org/synapse/pull/13534 Part of https://github.com/matrix-org/synapse/issues/13356 --- changelog.d/13541.misc | 1 + synapse/util/ratelimitutils.py | 43 ++++++++++++++++++++++++++++++++++++++---- 2 files changed, 40 insertions(+), 4 deletions(-) create mode 100644 changelog.d/13541.misc (limited to 'synapse/util') diff --git a/changelog.d/13541.misc b/changelog.d/13541.misc new file mode 100644 index 0000000000..b488bf74c3 --- /dev/null +++ b/changelog.d/13541.misc @@ -0,0 +1 @@ +Add metrics to track how the rate limiter is affecting requests (sleep/reject). diff --git a/synapse/util/ratelimitutils.py b/synapse/util/ratelimitutils.py index 434b02b97b..724d39b92f 100644 --- a/synapse/util/ratelimitutils.py +++ b/synapse/util/ratelimitutils.py @@ -30,7 +30,7 @@ from synapse.logging.context import ( run_in_background, ) from synapse.logging.opentracing import start_active_span -from synapse.metrics import Histogram +from synapse.metrics import Histogram, LaterGauge from synapse.util import Clock if typing.TYPE_CHECKING: @@ -74,6 +74,27 @@ class FederationRateLimiter: str, "_PerHostRatelimiter" ] = collections.defaultdict(new_limiter) + # We track the number of affected hosts per time-period so we can + # differentiate one really noisy homeserver from a general + # ratelimit tuning problem across the federation. + LaterGauge( + "synapse_rate_limit_sleep_affected_hosts", + "Number of hosts that had requests put to sleep", + [], + lambda: sum( + ratelimiter.should_sleep() for ratelimiter in self.ratelimiters.values() + ), + ) + LaterGauge( + "synapse_rate_limit_reject_affected_hosts", + "Number of hosts that had requests rejected", + [], + lambda: sum( + ratelimiter.should_reject() + for ratelimiter in self.ratelimiters.values() + ), + ) + def ratelimit(self, host: str) -> "_GeneratorContextManager[defer.Deferred[None]]": """Used to ratelimit an incoming request from a given host @@ -139,6 +160,21 @@ class _PerHostRatelimiter: finally: self._on_exit(request_id) + def should_reject(self) -> bool: + """ + Whether to reject the request if we already have too many queued up + (either sleeping or in the ready queue). + """ + queue_size = len(self.ready_request_queue) + len(self.sleeping_requests) + return queue_size > self.reject_limit + + def should_sleep(self) -> bool: + """ + Whether to sleep the request if we already have too many requests coming + through within the window. + """ + return len(self.request_times) > self.sleep_limit + def _on_enter(self, request_id: object) -> "defer.Deferred[None]": time_now = self.clock.time_msec() @@ -149,8 +185,7 @@ class _PerHostRatelimiter: # reject the request if we already have too many queued up (either # sleeping or in the ready queue). - queue_size = len(self.ready_request_queue) + len(self.sleeping_requests) - if queue_size > self.reject_limit: + if self.should_reject(): logger.debug("Ratelimiter(%s): rejecting request", self.host) rate_limit_reject_counter.inc() raise LimitExceededError( @@ -180,7 +215,7 @@ class _PerHostRatelimiter: len(self.request_times), ) - if len(self.request_times) > self.sleep_limit: + if self.should_sleep(): logger.debug( "Ratelimiter(%s) [%s]: sleeping request for %f sec", self.host, -- cgit 1.5.1 From b251cff8196e4130b2a6951c8fe569ed46779443 Mon Sep 17 00:00:00 2001 From: Sean Quah <8349537+squahtx@users.noreply.github.com> Date: Thu, 18 Aug 2022 16:26:26 +0100 Subject: Fix incorrect juggling of logging contexts in `_PerHostRatelimiter` (#13554) Signed-off-by: Sean Quah Co-authored-by: Richard van der Hoff <1389908+richvdh@users.noreply.github.com> --- changelog.d/13554.misc | 1 + synapse/util/ratelimitutils.py | 17 +++++++---------- 2 files changed, 8 insertions(+), 10 deletions(-) create mode 100644 changelog.d/13554.misc (limited to 'synapse/util') diff --git a/changelog.d/13554.misc b/changelog.d/13554.misc new file mode 100644 index 0000000000..99dbcebec8 --- /dev/null +++ b/changelog.d/13554.misc @@ -0,0 +1 @@ +Instrument `FederationStateIdsServlet` (`/state_ids`) for understandable traces in Jaeger. diff --git a/synapse/util/ratelimitutils.py b/synapse/util/ratelimitutils.py index 724d39b92f..f678b52cb4 100644 --- a/synapse/util/ratelimitutils.py +++ b/synapse/util/ratelimitutils.py @@ -154,7 +154,9 @@ class _PerHostRatelimiter: self.host = host request_id = object() - ret = self._on_enter(request_id) + # Ideally we'd use `Deferred.fromCoroutine()` here, to save on redundant + # type-checking, but we'd need Twisted >= 21.2. + ret = defer.ensureDeferred(self._on_enter_with_tracing(request_id)) try: yield ret finally: @@ -175,6 +177,10 @@ class _PerHostRatelimiter: """ return len(self.request_times) > self.sleep_limit + async def _on_enter_with_tracing(self, request_id: object) -> None: + with start_active_span("ratelimit wait"), queue_wait_timer.time(): + await self._on_enter(request_id) + def _on_enter(self, request_id: object) -> "defer.Deferred[None]": time_now = self.clock.time_msec() @@ -257,17 +263,8 @@ class _PerHostRatelimiter: # Ensure that we've properly cleaned up. self.sleeping_requests.discard(request_id) self.ready_request_queue.pop(request_id, None) - wait_span_scope.__exit__(None, None, None) - wait_timer_cm.__exit__(None, None, None) return r - # Tracing - wait_span_scope = start_active_span("ratelimit wait") - wait_span_scope.__enter__() - # Metrics - wait_timer_cm = queue_wait_timer.time() - wait_timer_cm.__enter__() - ret_defer.addCallbacks(on_start, on_err) ret_defer.addBoth(on_both) return make_deferred_yieldable(ret_defer) -- cgit 1.5.1 From 5e7847dc923142bc68834f9b9538ada3fdd887d5 Mon Sep 17 00:00:00 2001 From: Nick Mills-Barrett Date: Tue, 23 Aug 2022 10:49:59 +0100 Subject: Cache user IDs instead of profile objects (#13573) The profile objects are never used and increase cache size significantly. --- changelog.d/13573.misc | 1 + synapse/handlers/sync.py | 4 +- synapse/state/__init__.py | 13 +++--- synapse/storage/databases/main/roommember.py | 67 ++++++++++++---------------- synapse/util/caches/descriptors.py | 26 ++++++++--- 5 files changed, 57 insertions(+), 54 deletions(-) create mode 100644 changelog.d/13573.misc (limited to 'synapse/util') diff --git a/changelog.d/13573.misc b/changelog.d/13573.misc new file mode 100644 index 0000000000..1ce9c0c081 --- /dev/null +++ b/changelog.d/13573.misc @@ -0,0 +1 @@ +Cache user IDs instead of profiles to reduce cache memory usage. Contributed by Nick @ Beeper (@fizzadar). diff --git a/synapse/handlers/sync.py b/synapse/handlers/sync.py index b4d3f3958c..2d95b1fa24 100644 --- a/synapse/handlers/sync.py +++ b/synapse/handlers/sync.py @@ -2421,10 +2421,10 @@ class SyncHandler: joined_room.room_id, joined_room.event_pos.stream ) ) - users_in_room = await self.state.get_current_users_in_room( + user_ids_in_room = await self.state.get_current_user_ids_in_room( joined_room.room_id, extrems ) - if user_id in users_in_room: + if user_id in user_ids_in_room: joined_room_ids.add(joined_room.room_id) return frozenset(joined_room_ids) diff --git a/synapse/state/__init__.py b/synapse/state/__init__.py index c355e4f98a..3047e1b1ad 100644 --- a/synapse/state/__init__.py +++ b/synapse/state/__init__.py @@ -44,7 +44,6 @@ from synapse.logging.context import ContextResourceUsage from synapse.replication.http.state import ReplicationUpdateCurrentStateRestServlet from synapse.state import v1, v2 from synapse.storage.databases.main.events_worker import EventRedactBehaviour -from synapse.storage.roommember import ProfileInfo from synapse.storage.state import StateFilter from synapse.types import StateMap from synapse.util.async_helpers import Linearizer @@ -210,11 +209,11 @@ class StateHandler: ret = await self.resolve_state_groups_for_events(room_id, event_ids) return await ret.get_state(self._state_storage_controller, state_filter) - async def get_current_users_in_room( + async def get_current_user_ids_in_room( self, room_id: str, latest_event_ids: List[str] - ) -> Dict[str, ProfileInfo]: + ) -> Set[str]: """ - Get the users who are currently in a room. + Get the users IDs who are currently in a room. Note: This is much slower than using the equivalent method `DataStore.get_users_in_room` or `DataStore.get_users_in_room_with_profiles`, @@ -225,15 +224,15 @@ class StateHandler: room_id: The ID of the room. latest_event_ids: Precomputed list of latest event IDs. Will be computed if None. Returns: - Dictionary of user IDs to their profileinfo. + Set of user IDs in the room. """ assert latest_event_ids is not None - logger.debug("calling resolve_state_groups from get_current_users_in_room") + logger.debug("calling resolve_state_groups from get_current_user_ids_in_room") entry = await self.resolve_state_groups_for_events(room_id, latest_event_ids) state = await entry.get_state(self._state_storage_controller, StateFilter.all()) - return await self.store.get_joined_users_from_state(room_id, state, entry) + return await self.store.get_joined_user_ids_from_state(room_id, state, entry) async def get_hosts_in_room_at_events( self, room_id: str, event_ids: Collection[str] diff --git a/synapse/storage/databases/main/roommember.py b/synapse/storage/databases/main/roommember.py index 827c1f1efd..0eb024a809 100644 --- a/synapse/storage/databases/main/roommember.py +++ b/synapse/storage/databases/main/roommember.py @@ -835,9 +835,9 @@ class RoomMemberWorkerStore(EventsWorkerStore): return shared_room_ids or frozenset() - async def get_joined_users_from_state( + async def get_joined_user_ids_from_state( self, room_id: str, state: StateMap[str], state_entry: "_StateCacheEntry" - ) -> Dict[str, ProfileInfo]: + ) -> Set[str]: state_group: Union[object, int] = state_entry.state_group if not state_group: # If state_group is None it means it has yet to be assigned a @@ -848,25 +848,25 @@ class RoomMemberWorkerStore(EventsWorkerStore): assert state_group is not None with Measure(self._clock, "get_joined_users_from_state"): - return await self._get_joined_users_from_context( + return await self._get_joined_user_ids_from_context( room_id, state_group, state, context=state_entry ) @cached(num_args=2, iterable=True, max_entries=100000) - async def _get_joined_users_from_context( + async def _get_joined_user_ids_from_context( self, room_id: str, state_group: Union[object, int], current_state_ids: StateMap[str], event: Optional[EventBase] = None, context: Optional["_StateCacheEntry"] = None, - ) -> Dict[str, ProfileInfo]: + ) -> Set[str]: # We don't use `state_group`, it's there so that we can cache based # on it. However, it's important that it's never None, since two current_states # with a state_group of None are likely to be different. assert state_group is not None - users_in_room = {} + users_in_room = set() member_event_ids = [ e_id for key, e_id in current_state_ids.items() @@ -879,11 +879,11 @@ class RoomMemberWorkerStore(EventsWorkerStore): # If we do then we can reuse that result and simply update it with # any membership changes in `delta_ids` if context.prev_group and context.delta_ids: - prev_res = self._get_joined_users_from_context.cache.get_immediate( + prev_res = self._get_joined_user_ids_from_context.cache.get_immediate( (room_id, context.prev_group), None ) - if prev_res and isinstance(prev_res, dict): - users_in_room = dict(prev_res) + if prev_res and isinstance(prev_res, set): + users_in_room = prev_res member_event_ids = [ e_id for key, e_id in context.delta_ids.items() @@ -891,7 +891,7 @@ class RoomMemberWorkerStore(EventsWorkerStore): ] for etype, state_key in context.delta_ids: if etype == EventTypes.Member: - users_in_room.pop(state_key, None) + users_in_room.discard(state_key) # We check if we have any of the member event ids in the event cache # before we ask the DB @@ -908,42 +908,41 @@ class RoomMemberWorkerStore(EventsWorkerStore): ev_entry = event_map.get(event_id) if ev_entry and not ev_entry.event.rejected_reason: if ev_entry.event.membership == Membership.JOIN: - users_in_room[ev_entry.event.state_key] = ProfileInfo( - display_name=ev_entry.event.content.get("displayname", None), - avatar_url=ev_entry.event.content.get("avatar_url", None), - ) + users_in_room.add(ev_entry.event.state_key) else: missing_member_event_ids.append(event_id) if missing_member_event_ids: - event_to_memberships = await self._get_joined_profiles_from_event_ids( + event_to_memberships = await self._get_user_ids_from_membership_event_ids( missing_member_event_ids ) - users_in_room.update(row for row in event_to_memberships.values() if row) + users_in_room.update(event_to_memberships.values()) if event is not None and event.type == EventTypes.Member: if event.membership == Membership.JOIN: if event.event_id in member_event_ids: - users_in_room[event.state_key] = ProfileInfo( - display_name=event.content.get("displayname", None), - avatar_url=event.content.get("avatar_url", None), - ) + users_in_room.add(event.state_key) return users_in_room - @cached(max_entries=10000) - def _get_joined_profile_from_event_id( + @cached( + max_entries=10000, + # This name matches the old function that has been replaced - the cache name + # is kept here to maintain backwards compatibility. + name="_get_joined_profile_from_event_id", + ) + def _get_user_id_from_membership_event_id( self, event_id: str ) -> Optional[Tuple[str, ProfileInfo]]: raise NotImplementedError() @cachedList( - cached_method_name="_get_joined_profile_from_event_id", + cached_method_name="_get_user_id_from_membership_event_id", list_name="event_ids", ) - async def _get_joined_profiles_from_event_ids( + async def _get_user_ids_from_membership_event_ids( self, event_ids: Iterable[str] - ) -> Dict[str, Optional[Tuple[str, ProfileInfo]]]: + ) -> Dict[str, str]: """For given set of member event_ids check if they point to a join event and if so return the associated user and profile info. @@ -958,21 +957,13 @@ class RoomMemberWorkerStore(EventsWorkerStore): table="room_memberships", column="event_id", iterable=event_ids, - retcols=("user_id", "display_name", "avatar_url", "event_id"), + retcols=("user_id", "event_id"), keyvalues={"membership": Membership.JOIN}, batch_size=1000, - desc="_get_joined_profiles_from_event_ids", + desc="_get_user_ids_from_membership_event_ids", ) - return { - row["event_id"]: ( - row["user_id"], - ProfileInfo( - avatar_url=row["avatar_url"], display_name=row["display_name"] - ), - ) - for row in rows - } + return {row["event_id"]: row["user_id"] for row in rows} @cached(max_entries=10000) async def is_host_joined(self, room_id: str, host: str) -> bool: @@ -1131,12 +1122,12 @@ class RoomMemberWorkerStore(EventsWorkerStore): else: # The cache doesn't match the state group or prev state group, # so we calculate the result from first principles. - joined_users = await self.get_joined_users_from_state( + joined_user_ids = await self.get_joined_user_ids_from_state( room_id, state, state_entry ) cache.hosts_to_joined_users = {} - for user_id in joined_users: + for user_id in joined_user_ids: host = intern_string(get_domain_from_id(user_id)) cache.hosts_to_joined_users.setdefault(host, set()).add(user_id) diff --git a/synapse/util/caches/descriptors.py b/synapse/util/caches/descriptors.py index 867f315b2a..9d4bc89edb 100644 --- a/synapse/util/caches/descriptors.py +++ b/synapse/util/caches/descriptors.py @@ -73,8 +73,10 @@ class _CacheDescriptorBase: num_args: Optional[int], uncached_args: Optional[Collection[str]] = None, cache_context: bool = False, + name: Optional[str] = None, ): self.orig = orig + self.name = name or orig.__name__ arg_spec = inspect.getfullargspec(orig) all_args = arg_spec.args @@ -211,7 +213,7 @@ class LruCacheDescriptor(_CacheDescriptorBase): def __get__(self, obj: Optional[Any], owner: Optional[Type]) -> Callable[..., Any]: cache: LruCache[CacheKey, Any] = LruCache( - cache_name=self.orig.__name__, + cache_name=self.name, max_size=self.max_entries, ) @@ -241,7 +243,7 @@ class LruCacheDescriptor(_CacheDescriptorBase): wrapped = cast(_CachedFunction, _wrapped) wrapped.cache = cache - obj.__dict__[self.orig.__name__] = wrapped + obj.__dict__[self.name] = wrapped return wrapped @@ -301,12 +303,14 @@ class DeferredCacheDescriptor(_CacheDescriptorBase): cache_context: bool = False, iterable: bool = False, prune_unread_entries: bool = True, + name: Optional[str] = None, ): super().__init__( orig, num_args=num_args, uncached_args=uncached_args, cache_context=cache_context, + name=name, ) if tree and self.num_args < 2: @@ -321,7 +325,7 @@ class DeferredCacheDescriptor(_CacheDescriptorBase): def __get__(self, obj: Optional[Any], owner: Optional[Type]) -> Callable[..., Any]: cache: DeferredCache[CacheKey, Any] = DeferredCache( - name=self.orig.__name__, + name=self.name, max_entries=self.max_entries, tree=self.tree, iterable=self.iterable, @@ -372,7 +376,7 @@ class DeferredCacheDescriptor(_CacheDescriptorBase): wrapped.cache = cache wrapped.num_args = self.num_args - obj.__dict__[self.orig.__name__] = wrapped + obj.__dict__[self.name] = wrapped return wrapped @@ -393,6 +397,7 @@ class DeferredCacheListDescriptor(_CacheDescriptorBase): cached_method_name: str, list_name: str, num_args: Optional[int] = None, + name: Optional[str] = None, ): """ Args: @@ -403,7 +408,7 @@ class DeferredCacheListDescriptor(_CacheDescriptorBase): but including list_name) to use as cache keys. Defaults to all named args of the function. """ - super().__init__(orig, num_args=num_args, uncached_args=None) + super().__init__(orig, num_args=num_args, uncached_args=None, name=name) self.list_name = list_name @@ -525,7 +530,7 @@ class DeferredCacheListDescriptor(_CacheDescriptorBase): else: return defer.succeed(results) - obj.__dict__[self.orig.__name__] = wrapped + obj.__dict__[self.name] = wrapped return wrapped @@ -577,6 +582,7 @@ def cached( cache_context: bool = False, iterable: bool = False, prune_unread_entries: bool = True, + name: Optional[str] = None, ) -> Callable[[F], _CachedFunction[F]]: func = lambda orig: DeferredCacheDescriptor( orig, @@ -587,13 +593,18 @@ def cached( cache_context=cache_context, iterable=iterable, prune_unread_entries=prune_unread_entries, + name=name, ) return cast(Callable[[F], _CachedFunction[F]], func) def cachedList( - *, cached_method_name: str, list_name: str, num_args: Optional[int] = None + *, + cached_method_name: str, + list_name: str, + num_args: Optional[int] = None, + name: Optional[str] = None, ) -> Callable[[F], _CachedFunction[F]]: """Creates a descriptor that wraps a function in a `DeferredCacheListDescriptor`. @@ -628,6 +639,7 @@ def cachedList( cached_method_name=cached_method_name, list_name=list_name, num_args=num_args, + name=name, ) return cast(Callable[[F], _CachedFunction[F]], func) -- cgit 1.5.1 From f7ddfe17a30a50205a23bf5ca4d7d71e691e1e48 Mon Sep 17 00:00:00 2001 From: Erik Johnston Date: Tue, 23 Aug 2022 15:53:27 +0100 Subject: Speed up `@cachedList` (#13591) This speeds things up by ~2x. The vast majority of the time is now spent in `LruCache` moving things around the linked lists. We do this via two things: 1. Don't create a deferred per-key during bulk set operations in `DeferredCache`. Instead, only create them if a subsequent caller asks for the key. 2. Add a bulk lookup API to `DeferredCache` rather than use a loop. --- changelog.d/13591.misc | 1 + synapse/util/caches/deferred_cache.py | 346 +++++++++++++++++++++++++--------- synapse/util/caches/descriptors.py | 89 ++++----- synapse/util/caches/treecache.py | 3 + 4 files changed, 298 insertions(+), 141 deletions(-) create mode 100644 changelog.d/13591.misc (limited to 'synapse/util') diff --git a/changelog.d/13591.misc b/changelog.d/13591.misc new file mode 100644 index 0000000000..080e865e55 --- /dev/null +++ b/changelog.d/13591.misc @@ -0,0 +1 @@ +Improve performance of `@cachedList`. diff --git a/synapse/util/caches/deferred_cache.py b/synapse/util/caches/deferred_cache.py index 1d6ec22191..6425f851ea 100644 --- a/synapse/util/caches/deferred_cache.py +++ b/synapse/util/caches/deferred_cache.py @@ -14,15 +14,19 @@ # See the License for the specific language governing permissions and # limitations under the License. +import abc import enum import threading from typing import ( Callable, + Collection, + Dict, Generic, - Iterable, MutableMapping, Optional, + Set, Sized, + Tuple, TypeVar, Union, cast, @@ -31,7 +35,6 @@ from typing import ( from prometheus_client import Gauge from twisted.internet import defer -from twisted.python import failure from twisted.python.failure import Failure from synapse.util.async_helpers import ObservableDeferred @@ -94,7 +97,7 @@ class DeferredCache(Generic[KT, VT]): # _pending_deferred_cache maps from the key value to a `CacheEntry` object. self._pending_deferred_cache: Union[ - TreeCache, "MutableMapping[KT, CacheEntry]" + TreeCache, "MutableMapping[KT, CacheEntry[KT, VT]]" ] = cache_type() def metrics_cb() -> None: @@ -159,15 +162,16 @@ class DeferredCache(Generic[KT, VT]): Raises: KeyError if the key is not found in the cache """ - callbacks = [callback] if callback else [] val = self._pending_deferred_cache.get(key, _Sentinel.sentinel) if val is not _Sentinel.sentinel: - val.callbacks.update(callbacks) + val.add_invalidation_callback(key, callback) if update_metrics: m = self.cache.metrics assert m # we always have a name, so should always have metrics m.inc_hits() - return val.deferred.observe() + return val.deferred(key) + + callbacks = (callback,) if callback else () val2 = self.cache.get( key, _Sentinel.sentinel, callbacks=callbacks, update_metrics=update_metrics @@ -177,6 +181,73 @@ class DeferredCache(Generic[KT, VT]): else: return defer.succeed(val2) + def get_bulk( + self, + keys: Collection[KT], + callback: Optional[Callable[[], None]] = None, + ) -> Tuple[Dict[KT, VT], Optional["defer.Deferred[Dict[KT, VT]]"], Collection[KT]]: + """Bulk lookup of items in the cache. + + Returns: + A 3-tuple of: + 1. a dict of key/value of items already cached; + 2. a deferred that resolves to a dict of key/value of items + we're already fetching; and + 3. a collection of keys that don't appear in the previous two. + """ + + # The cached results + cached = {} + + # List of pending deferreds + pending = [] + + # Dict that gets filled out when the pending deferreds complete + pending_results = {} + + # List of keys that aren't in either cache + missing = [] + + callbacks = (callback,) if callback else () + + for key in keys: + # Check if its in the main cache. + immediate_value = self.cache.get( + key, + _Sentinel.sentinel, + callbacks=callbacks, + ) + if immediate_value is not _Sentinel.sentinel: + cached[key] = immediate_value + continue + + # Check if its in the pending cache + pending_value = self._pending_deferred_cache.get(key, _Sentinel.sentinel) + if pending_value is not _Sentinel.sentinel: + pending_value.add_invalidation_callback(key, callback) + + def completed_cb(value: VT, key: KT) -> VT: + pending_results[key] = value + return value + + # Add a callback to fill out `pending_results` when that completes + d = pending_value.deferred(key).addCallback(completed_cb, key) + pending.append(d) + continue + + # Not in either cache + missing.append(key) + + # If we've got pending deferreds, squash them into a single one that + # returns `pending_results`. + pending_deferred = None + if pending: + pending_deferred = defer.gatherResults( + pending, consumeErrors=True + ).addCallback(lambda _: pending_results) + + return (cached, pending_deferred, missing) + def get_immediate( self, key: KT, default: T, update_metrics: bool = True ) -> Union[VT, T]: @@ -218,84 +289,89 @@ class DeferredCache(Generic[KT, VT]): value: a deferred which will complete with a result to add to the cache callback: An optional callback to be called when the entry is invalidated """ - if not isinstance(value, defer.Deferred): - raise TypeError("not a Deferred") - - callbacks = [callback] if callback else [] self.check_thread() - existing_entry = self._pending_deferred_cache.pop(key, None) - if existing_entry: - existing_entry.invalidate() + self._pending_deferred_cache.pop(key, None) # XXX: why don't we invalidate the entry in `self.cache` yet? - # we can save a whole load of effort if the deferred is ready. - if value.called: - result = value.result - if not isinstance(result, failure.Failure): - self.cache.set(key, cast(VT, result), callbacks) - return value - # otherwise, we'll add an entry to the _pending_deferred_cache for now, # and add callbacks to add it to the cache properly later. + entry = CacheEntrySingle[KT, VT](value) + entry.add_invalidation_callback(key, callback) + self._pending_deferred_cache[key] = entry + deferred = entry.deferred(key).addCallbacks( + self._completed_callback, + self._error_callback, + callbackArgs=(entry, key), + errbackArgs=(entry, key), + ) - observable = ObservableDeferred(value, consumeErrors=True) - observer = observable.observe() - entry = CacheEntry(deferred=observable, callbacks=callbacks) + # we return a new Deferred which will be called before any subsequent observers. + return deferred - self._pending_deferred_cache[key] = entry + def start_bulk_input( + self, + keys: Collection[KT], + callback: Optional[Callable[[], None]] = None, + ) -> "CacheMultipleEntries[KT, VT]": + """Bulk set API for use when fetching multiple keys at once from the DB. - def compare_and_pop() -> bool: - """Check if our entry is still the one in _pending_deferred_cache, and - if so, pop it. - - Returns true if the entries matched. - """ - existing_entry = self._pending_deferred_cache.pop(key, None) - if existing_entry is entry: - return True - - # oops, the _pending_deferred_cache has been updated since - # we started our query, so we are out of date. - # - # Better put back whatever we took out. (We do it this way - # round, rather than peeking into the _pending_deferred_cache - # and then removing on a match, to make the common case faster) - if existing_entry is not None: - self._pending_deferred_cache[key] = existing_entry - - return False - - def cb(result: VT) -> None: - if compare_and_pop(): - self.cache.set(key, result, entry.callbacks) - else: - # we're not going to put this entry into the cache, so need - # to make sure that the invalidation callbacks are called. - # That was probably done when _pending_deferred_cache was - # updated, but it's possible that `set` was called without - # `invalidate` being previously called, in which case it may - # not have been. Either way, let's double-check now. - entry.invalidate() - - def eb(_fail: Failure) -> None: - compare_and_pop() - entry.invalidate() - - # once the deferred completes, we can move the entry from the - # _pending_deferred_cache to the real cache. - # - observer.addCallbacks(cb, eb) + Called *before* starting the fetch from the DB, and the caller *must* + call either `complete_bulk(..)` or `error_bulk(..)` on the return value. + """ - # we return a new Deferred which will be called before any subsequent observers. - return observable.observe() + entry = CacheMultipleEntries[KT, VT]() + entry.add_global_invalidation_callback(callback) + + for key in keys: + self._pending_deferred_cache[key] = entry + + return entry + + def _completed_callback( + self, value: VT, entry: "CacheEntry[KT, VT]", key: KT + ) -> VT: + """Called when a deferred is completed.""" + # We check if the current entry matches the entry associated with the + # deferred. If they don't match then it got invalidated. + current_entry = self._pending_deferred_cache.pop(key, None) + if current_entry is not entry: + if current_entry: + self._pending_deferred_cache[key] = current_entry + return value + + self.cache.set(key, value, entry.get_invalidation_callbacks(key)) + + return value + + def _error_callback( + self, + failure: Failure, + entry: "CacheEntry[KT, VT]", + key: KT, + ) -> Failure: + """Called when a deferred errors.""" + + # We check if the current entry matches the entry associated with the + # deferred. If they don't match then it got invalidated. + current_entry = self._pending_deferred_cache.pop(key, None) + if current_entry is not entry: + if current_entry: + self._pending_deferred_cache[key] = current_entry + return failure + + for cb in entry.get_invalidation_callbacks(key): + cb() + + return failure def prefill( self, key: KT, value: VT, callback: Optional[Callable[[], None]] = None ) -> None: - callbacks = [callback] if callback else [] + callbacks = (callback,) if callback else () self.cache.set(key, value, callbacks=callbacks) + self._pending_deferred_cache.pop(key, None) def invalidate(self, key: KT) -> None: """Delete a key, or tree of entries @@ -311,41 +387,129 @@ class DeferredCache(Generic[KT, VT]): self.cache.del_multi(key) # if we have a pending lookup for this key, remove it from the - # _pending_deferred_cache, which will (a) stop it being returned - # for future queries and (b) stop it being persisted as a proper entry + # _pending_deferred_cache, which will (a) stop it being returned for + # future queries and (b) stop it being persisted as a proper entry # in self.cache. entry = self._pending_deferred_cache.pop(key, None) - - # run the invalidation callbacks now, rather than waiting for the - # deferred to resolve. if entry: # _pending_deferred_cache.pop should either return a CacheEntry, or, in the # case of a TreeCache, a dict of keys to cache entries. Either way calling # iterate_tree_cache_entry on it will do the right thing. for entry in iterate_tree_cache_entry(entry): - entry.invalidate() + for cb in entry.get_invalidation_callbacks(key): + cb() def invalidate_all(self) -> None: self.check_thread() self.cache.clear() - for entry in self._pending_deferred_cache.values(): - entry.invalidate() + for key, entry in self._pending_deferred_cache.items(): + for cb in entry.get_invalidation_callbacks(key): + cb() + self._pending_deferred_cache.clear() -class CacheEntry: - __slots__ = ["deferred", "callbacks", "invalidated"] +class CacheEntry(Generic[KT, VT], metaclass=abc.ABCMeta): + """Abstract class for entries in `DeferredCache[KT, VT]`""" - def __init__( - self, deferred: ObservableDeferred, callbacks: Iterable[Callable[[], None]] - ): - self.deferred = deferred - self.callbacks = set(callbacks) - self.invalidated = False - - def invalidate(self) -> None: - if not self.invalidated: - self.invalidated = True - for callback in self.callbacks: - callback() - self.callbacks.clear() + @abc.abstractmethod + def deferred(self, key: KT) -> "defer.Deferred[VT]": + """Get a deferred that a caller can wait on to get the value at the + given key""" + ... + + @abc.abstractmethod + def add_invalidation_callback( + self, key: KT, callback: Optional[Callable[[], None]] + ) -> None: + """Add an invalidation callback""" + ... + + @abc.abstractmethod + def get_invalidation_callbacks(self, key: KT) -> Collection[Callable[[], None]]: + """Get all invalidation callbacks""" + ... + + +class CacheEntrySingle(CacheEntry[KT, VT]): + """An implementation of `CacheEntry` wrapping a deferred that results in a + single cache entry. + """ + + __slots__ = ["_deferred", "_callbacks"] + + def __init__(self, deferred: "defer.Deferred[VT]") -> None: + self._deferred = ObservableDeferred(deferred, consumeErrors=True) + self._callbacks: Set[Callable[[], None]] = set() + + def deferred(self, key: KT) -> "defer.Deferred[VT]": + return self._deferred.observe() + + def add_invalidation_callback( + self, key: KT, callback: Optional[Callable[[], None]] + ) -> None: + if callback is None: + return + + self._callbacks.add(callback) + + def get_invalidation_callbacks(self, key: KT) -> Collection[Callable[[], None]]: + return self._callbacks + + +class CacheMultipleEntries(CacheEntry[KT, VT]): + """Cache entry that is used for bulk lookups and insertions.""" + + __slots__ = ["_deferred", "_callbacks", "_global_callbacks"] + + def __init__(self) -> None: + self._deferred: Optional[ObservableDeferred[Dict[KT, VT]]] = None + self._callbacks: Dict[KT, Set[Callable[[], None]]] = {} + self._global_callbacks: Set[Callable[[], None]] = set() + + def deferred(self, key: KT) -> "defer.Deferred[VT]": + if not self._deferred: + self._deferred = ObservableDeferred(defer.Deferred(), consumeErrors=True) + return self._deferred.observe().addCallback(lambda res: res.get(key)) + + def add_invalidation_callback( + self, key: KT, callback: Optional[Callable[[], None]] + ) -> None: + if callback is None: + return + + self._callbacks.setdefault(key, set()).add(callback) + + def get_invalidation_callbacks(self, key: KT) -> Collection[Callable[[], None]]: + return self._callbacks.get(key, set()) | self._global_callbacks + + def add_global_invalidation_callback( + self, callback: Optional[Callable[[], None]] + ) -> None: + """Add a callback for when any keys get invalidated.""" + if callback is None: + return + + self._global_callbacks.add(callback) + + def complete_bulk( + self, + cache: DeferredCache[KT, VT], + result: Dict[KT, VT], + ) -> None: + """Called when there is a result""" + for key, value in result.items(): + cache._completed_callback(value, self, key) + + if self._deferred: + self._deferred.callback(result) + + def error_bulk( + self, cache: DeferredCache[KT, VT], keys: Collection[KT], failure: Failure + ) -> None: + """Called when bulk lookup failed.""" + for key in keys: + cache._error_callback(failure, self, key) + + if self._deferred: + self._deferred.errback(failure) diff --git a/synapse/util/caches/descriptors.py b/synapse/util/caches/descriptors.py index 9d4bc89edb..10aff4d04a 100644 --- a/synapse/util/caches/descriptors.py +++ b/synapse/util/caches/descriptors.py @@ -25,6 +25,7 @@ from typing import ( Generic, Hashable, Iterable, + List, Mapping, Optional, Sequence, @@ -440,16 +441,6 @@ class DeferredCacheListDescriptor(_CacheDescriptorBase): keyargs = [arg_dict[arg_nm] for arg_nm in self.arg_names] list_args = arg_dict[self.list_name] - results = {} - - def update_results_dict(res: Any, arg: Hashable) -> None: - results[arg] = res - - # list of deferreds to wait for - cached_defers = [] - - missing = set() - # If the cache takes a single arg then that is used as the key, # otherwise a tuple is used. if num_args == 1: @@ -457,6 +448,9 @@ class DeferredCacheListDescriptor(_CacheDescriptorBase): def arg_to_cache_key(arg: Hashable) -> Hashable: return arg + def cache_key_to_arg(key: tuple) -> Hashable: + return key + else: keylist = list(keyargs) @@ -464,58 +458,53 @@ class DeferredCacheListDescriptor(_CacheDescriptorBase): keylist[self.list_pos] = arg return tuple(keylist) - for arg in list_args: - try: - res = cache.get(arg_to_cache_key(arg), callback=invalidate_callback) - if not res.called: - res.addCallback(update_results_dict, arg) - cached_defers.append(res) - else: - results[arg] = res.result - except KeyError: - missing.add(arg) + def cache_key_to_arg(key: tuple) -> Hashable: + return key[self.list_pos] + + cache_keys = [arg_to_cache_key(arg) for arg in list_args] + immediate_results, pending_deferred, missing = cache.get_bulk( + cache_keys, callback=invalidate_callback + ) + + results = {cache_key_to_arg(key): v for key, v in immediate_results.items()} + + cached_defers: List["defer.Deferred[Any]"] = [] + if pending_deferred: + + def update_results(r: Dict) -> None: + for k, v in r.items(): + results[cache_key_to_arg(k)] = v + + pending_deferred.addCallback(update_results) + cached_defers.append(pending_deferred) if missing: - # we need a deferred for each entry in the list, - # which we put in the cache. Each deferred resolves with the - # relevant result for that key. - deferreds_map = {} - for arg in missing: - deferred: "defer.Deferred[Any]" = defer.Deferred() - deferreds_map[arg] = deferred - key = arg_to_cache_key(arg) - cached_defers.append( - cache.set(key, deferred, callback=invalidate_callback) - ) + cache_entry = cache.start_bulk_input(missing, invalidate_callback) def complete_all(res: Dict[Hashable, Any]) -> None: - # the wrapped function has completed. It returns a dict. - # We can now update our own result map, and then resolve the - # observable deferreds in the cache. - for e, d1 in deferreds_map.items(): - val = res.get(e, None) - # make sure we update the results map before running the - # deferreds, because as soon as we run the last deferred, the - # gatherResults() below will complete and return the result - # dict to our caller. - results[e] = val - d1.callback(val) + missing_results = {} + for key in missing: + arg = cache_key_to_arg(key) + val = res.get(arg, None) + + results[arg] = val + missing_results[key] = val + + cache_entry.complete_bulk(cache, missing_results) def errback_all(f: Failure) -> None: - # the wrapped function has failed. Propagate the failure into - # the cache, which will invalidate the entry, and cause the - # relevant cached_deferreds to fail, which will propagate the - # failure to our caller. - for d1 in deferreds_map.values(): - d1.errback(f) + cache_entry.error_bulk(cache, missing, f) args_to_call = dict(arg_dict) - args_to_call[self.list_name] = missing + args_to_call[self.list_name] = { + cache_key_to_arg(key) for key in missing + } # dispatch the call, and attach the two handlers - defer.maybeDeferred( + missing_d = defer.maybeDeferred( preserve_fn(self.orig), **args_to_call ).addCallbacks(complete_all, errback_all) + cached_defers.append(missing_d) if cached_defers: d = defer.gatherResults(cached_defers, consumeErrors=True).addCallbacks( diff --git a/synapse/util/caches/treecache.py b/synapse/util/caches/treecache.py index c1b8ec0c73..fec31da2b6 100644 --- a/synapse/util/caches/treecache.py +++ b/synapse/util/caches/treecache.py @@ -135,6 +135,9 @@ class TreeCache: def values(self): return iterate_tree_cache_entry(self.root) + def items(self): + return iterate_tree_cache_items((), self.root) + def __len__(self) -> int: return self.size -- cgit 1.5.1 From be4250c7a888e314e361df42042bfa344ab65d55 Mon Sep 17 00:00:00 2001 From: reivilibre Date: Wed, 24 Aug 2022 11:35:54 +0000 Subject: Add experimental configuration option to allow disabling legacy Prometheus metric names. (#13540) Co-authored-by: David Robertson --- changelog.d/13540.misc | 1 + synapse/app/_base.py | 39 ++++- synapse/app/generic_worker.py | 6 +- synapse/app/homeserver.py | 6 +- synapse/config/metrics.py | 29 ++++ synapse/metrics/__init__.py | 4 +- synapse/metrics/_exposition.py | 262 ------------------------------- synapse/metrics/_legacy_exposition.py | 284 ++++++++++++++++++++++++++++++++++ synapse/util/caches/__init__.py | 16 +- tests/test_metrics.py | 36 +++++ 10 files changed, 406 insertions(+), 277 deletions(-) create mode 100644 changelog.d/13540.misc delete mode 100644 synapse/metrics/_exposition.py create mode 100644 synapse/metrics/_legacy_exposition.py (limited to 'synapse/util') diff --git a/changelog.d/13540.misc b/changelog.d/13540.misc new file mode 100644 index 0000000000..07ace50b12 --- /dev/null +++ b/changelog.d/13540.misc @@ -0,0 +1 @@ +Add experimental configuration option to allow disabling legacy Prometheus metric names. \ No newline at end of file diff --git a/synapse/app/_base.py b/synapse/app/_base.py index 923891ae0d..4742435d3b 100644 --- a/synapse/app/_base.py +++ b/synapse/app/_base.py @@ -266,15 +266,48 @@ def register_start( reactor.callWhenRunning(lambda: defer.ensureDeferred(wrapper())) -def listen_metrics(bind_addresses: Iterable[str], port: int) -> None: +def listen_metrics( + bind_addresses: Iterable[str], port: int, enable_legacy_metric_names: bool +) -> None: """ Start Prometheus metrics server. """ - from synapse.metrics import RegistryProxy, start_http_server + from prometheus_client import start_http_server as start_http_server_prometheus + + from synapse.metrics import ( + RegistryProxy, + start_http_server as start_http_server_legacy, + ) for host in bind_addresses: logger.info("Starting metrics listener on %s:%d", host, port) - start_http_server(port, addr=host, registry=RegistryProxy) + if enable_legacy_metric_names: + start_http_server_legacy(port, addr=host, registry=RegistryProxy) + else: + _set_prometheus_client_use_created_metrics(False) + start_http_server_prometheus(port, addr=host, registry=RegistryProxy) + + +def _set_prometheus_client_use_created_metrics(new_value: bool) -> None: + """ + Sets whether prometheus_client should expose `_created`-suffixed metrics for + all gauges, histograms and summaries. + There is no programmatic way to disable this without poking at internals; + the proper way is to use an environment variable which prometheus_client + loads at import time. + + The motivation for disabling these `_created` metrics is that they're + a waste of space as they're not useful but they take up space in Prometheus. + """ + + import prometheus_client.metrics + + if hasattr(prometheus_client.metrics, "_use_created"): + prometheus_client.metrics._use_created = new_value + else: + logger.error( + "Can't disable `_created` metrics in prometheus_client (brittle hack broken?)" + ) def listen_manhole( diff --git a/synapse/app/generic_worker.py b/synapse/app/generic_worker.py index 30e21d9707..5e3825fca6 100644 --- a/synapse/app/generic_worker.py +++ b/synapse/app/generic_worker.py @@ -412,7 +412,11 @@ class GenericWorkerServer(HomeServer): "enable_metrics is not True!" ) else: - _base.listen_metrics(listener.bind_addresses, listener.port) + _base.listen_metrics( + listener.bind_addresses, + listener.port, + enable_legacy_metric_names=self.config.metrics.enable_legacy_metrics, + ) else: logger.warning("Unsupported listener type: %s", listener.type) diff --git a/synapse/app/homeserver.py b/synapse/app/homeserver.py index 68993d91a9..e57a926032 100644 --- a/synapse/app/homeserver.py +++ b/synapse/app/homeserver.py @@ -307,7 +307,11 @@ class SynapseHomeServer(HomeServer): "enable_metrics is not True!" ) else: - _base.listen_metrics(listener.bind_addresses, listener.port) + _base.listen_metrics( + listener.bind_addresses, + listener.port, + enable_legacy_metric_names=self.config.metrics.enable_legacy_metrics, + ) else: # this shouldn't happen, as the listener type should have been checked # during parsing diff --git a/synapse/config/metrics.py b/synapse/config/metrics.py index 3b42be5b5b..f3134834e5 100644 --- a/synapse/config/metrics.py +++ b/synapse/config/metrics.py @@ -42,6 +42,35 @@ class MetricsConfig(Config): def read_config(self, config: JsonDict, **kwargs: Any) -> None: self.enable_metrics = config.get("enable_metrics", False) + + """ + ### `enable_legacy_metrics` (experimental) + + **Experimental: this option may be removed or have its behaviour + changed at any time, with no notice.** + + Set to `true` to publish both legacy and non-legacy Prometheus metric names, + or to `false` to only publish non-legacy Prometheus metric names. + Defaults to `true`. Has no effect if `enable_metrics` is `false`. + + Legacy metric names include: + - metrics containing colons in the name, such as `synapse_util_caches_response_cache:hits`, because colons are supposed to be reserved for user-defined recording rules; + - counters that don't end with the `_total` suffix, such as `synapse_federation_client_sent_edus`, therefore not adhering to the OpenMetrics standard. + + These legacy metric names are unconventional and not compliant with OpenMetrics standards. + They are included for backwards compatibility. + + Example configuration: + ```yaml + enable_legacy_metrics: false + ``` + + See https://github.com/matrix-org/synapse/issues/11106 for context. + + *Since v1.67.0.* + """ + self.enable_legacy_metrics = config.get("enable_legacy_metrics", True) + self.report_stats = config.get("report_stats", None) self.report_stats_endpoint = config.get( "report_stats_endpoint", "https://matrix.org/report-usage-stats/push" diff --git a/synapse/metrics/__init__.py b/synapse/metrics/__init__.py index 496fce2ecc..c3d3daf877 100644 --- a/synapse/metrics/__init__.py +++ b/synapse/metrics/__init__.py @@ -46,12 +46,12 @@ from twisted.python.threadpool import ThreadPool # This module is imported for its side effects; flake8 needn't warn that it's unused. import synapse.metrics._reactor_metrics # noqa: F401 -from synapse.metrics._exposition import ( +from synapse.metrics._gc import MIN_TIME_BETWEEN_GCS, install_gc_manager +from synapse.metrics._legacy_exposition import ( MetricsResource, generate_latest, start_http_server, ) -from synapse.metrics._gc import MIN_TIME_BETWEEN_GCS, install_gc_manager from synapse.metrics._types import Collector from synapse.util import SYNAPSE_VERSION diff --git a/synapse/metrics/_exposition.py b/synapse/metrics/_exposition.py deleted file mode 100644 index 353d0a63b6..0000000000 --- a/synapse/metrics/_exposition.py +++ /dev/null @@ -1,262 +0,0 @@ -# Copyright 2015-2019 Prometheus Python Client Developers -# Copyright 2019 Matrix.org Foundation C.I.C. -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -""" -This code is based off `prometheus_client/exposition.py` from version 0.7.1. - -Due to the renaming of metrics in prometheus_client 0.4.0, this customised -vendoring of the code will emit both the old versions that Synapse dashboards -expect, and the newer "best practice" version of the up-to-date official client. -""" - -import math -import threading -from http.server import BaseHTTPRequestHandler, HTTPServer -from socketserver import ThreadingMixIn -from typing import Any, Dict, List, Type, Union -from urllib.parse import parse_qs, urlparse - -from prometheus_client import REGISTRY, CollectorRegistry -from prometheus_client.core import Sample - -from twisted.web.resource import Resource -from twisted.web.server import Request - -from synapse.util import caches - -CONTENT_TYPE_LATEST = "text/plain; version=0.0.4; charset=utf-8" - - -def floatToGoString(d: Union[int, float]) -> str: - d = float(d) - if d == math.inf: - return "+Inf" - elif d == -math.inf: - return "-Inf" - elif math.isnan(d): - return "NaN" - else: - s = repr(d) - dot = s.find(".") - # Go switches to exponents sooner than Python. - # We only need to care about positive values for le/quantile. - if d > 0 and dot > 6: - mantissa = f"{s[0]}.{s[1:dot]}{s[dot + 1 :]}".rstrip("0.") - return f"{mantissa}e+0{dot - 1}" - return s - - -def sample_line(line: Sample, name: str) -> str: - if line.labels: - labelstr = "{{{0}}}".format( - ",".join( - [ - '{}="{}"'.format( - k, - v.replace("\\", r"\\").replace("\n", r"\n").replace('"', r"\""), - ) - for k, v in sorted(line.labels.items()) - ] - ) - ) - else: - labelstr = "" - timestamp = "" - if line.timestamp is not None: - # Convert to milliseconds. - timestamp = f" {int(float(line.timestamp) * 1000):d}" - return "{}{} {}{}\n".format(name, labelstr, floatToGoString(line.value), timestamp) - - -def generate_latest(registry: CollectorRegistry, emit_help: bool = False) -> bytes: - - # Trigger the cache metrics to be rescraped, which updates the common - # metrics but do not produce metrics themselves - for collector in caches.collectors_by_name.values(): - collector.collect() - - output = [] - - for metric in registry.collect(): - if not metric.samples: - # No samples, don't bother. - continue - - mname = metric.name - mnewname = metric.name - mtype = metric.type - - # OpenMetrics -> Prometheus - if mtype == "counter": - mnewname = mnewname + "_total" - elif mtype == "info": - mtype = "gauge" - mnewname = mnewname + "_info" - elif mtype == "stateset": - mtype = "gauge" - elif mtype == "gaugehistogram": - mtype = "histogram" - elif mtype == "unknown": - mtype = "untyped" - - # Output in the old format for compatibility. - if emit_help: - output.append( - "# HELP {} {}\n".format( - mname, - metric.documentation.replace("\\", r"\\").replace("\n", r"\n"), - ) - ) - output.append(f"# TYPE {mname} {mtype}\n") - - om_samples: Dict[str, List[str]] = {} - for s in metric.samples: - for suffix in ["_created", "_gsum", "_gcount"]: - if s.name == metric.name + suffix: - # OpenMetrics specific sample, put in a gauge at the end. - # (these come from gaugehistograms which don't get renamed, - # so no need to faff with mnewname) - om_samples.setdefault(suffix, []).append(sample_line(s, s.name)) - break - else: - newname = s.name.replace(mnewname, mname) - if ":" in newname and newname.endswith("_total"): - newname = newname[: -len("_total")] - output.append(sample_line(s, newname)) - - for suffix, lines in sorted(om_samples.items()): - if emit_help: - output.append( - "# HELP {}{} {}\n".format( - metric.name, - suffix, - metric.documentation.replace("\\", r"\\").replace("\n", r"\n"), - ) - ) - output.append(f"# TYPE {metric.name}{suffix} gauge\n") - output.extend(lines) - - # Get rid of the weird colon things while we're at it - if mtype == "counter": - mnewname = mnewname.replace(":total", "") - mnewname = mnewname.replace(":", "_") - - if mname == mnewname: - continue - - # Also output in the new format, if it's different. - if emit_help: - output.append( - "# HELP {} {}\n".format( - mnewname, - metric.documentation.replace("\\", r"\\").replace("\n", r"\n"), - ) - ) - output.append(f"# TYPE {mnewname} {mtype}\n") - - for s in metric.samples: - # Get rid of the OpenMetrics specific samples (we should already have - # dealt with them above anyway.) - for suffix in ["_created", "_gsum", "_gcount"]: - if s.name == metric.name + suffix: - break - else: - output.append( - sample_line(s, s.name.replace(":total", "").replace(":", "_")) - ) - - return "".join(output).encode("utf-8") - - -class MetricsHandler(BaseHTTPRequestHandler): - """HTTP handler that gives metrics from ``REGISTRY``.""" - - registry = REGISTRY - - def do_GET(self) -> None: - registry = self.registry - params = parse_qs(urlparse(self.path).query) - - if "help" in params: - emit_help = True - else: - emit_help = False - - try: - output = generate_latest(registry, emit_help=emit_help) - except Exception: - self.send_error(500, "error generating metric output") - raise - self.send_response(200) - self.send_header("Content-Type", CONTENT_TYPE_LATEST) - self.send_header("Content-Length", str(len(output))) - self.end_headers() - self.wfile.write(output) - - def log_message(self, format: str, *args: Any) -> None: - """Log nothing.""" - - @classmethod - def factory(cls, registry: CollectorRegistry) -> Type: - """Returns a dynamic MetricsHandler class tied - to the passed registry. - """ - # This implementation relies on MetricsHandler.registry - # (defined above and defaulted to REGISTRY). - - # As we have unicode_literals, we need to create a str() - # object for type(). - cls_name = str(cls.__name__) - MyMetricsHandler = type(cls_name, (cls, object), {"registry": registry}) - return MyMetricsHandler - - -class _ThreadingSimpleServer(ThreadingMixIn, HTTPServer): - """Thread per request HTTP server.""" - - # Make worker threads "fire and forget". Beginning with Python 3.7 this - # prevents a memory leak because ``ThreadingMixIn`` starts to gather all - # non-daemon threads in a list in order to join on them at server close. - # Enabling daemon threads virtually makes ``_ThreadingSimpleServer`` the - # same as Python 3.7's ``ThreadingHTTPServer``. - daemon_threads = True - - -def start_http_server( - port: int, addr: str = "", registry: CollectorRegistry = REGISTRY -) -> None: - """Starts an HTTP server for prometheus metrics as a daemon thread""" - CustomMetricsHandler = MetricsHandler.factory(registry) - httpd = _ThreadingSimpleServer((addr, port), CustomMetricsHandler) - t = threading.Thread(target=httpd.serve_forever) - t.daemon = True - t.start() - - -class MetricsResource(Resource): - """ - Twisted ``Resource`` that serves prometheus metrics. - """ - - isLeaf = True - - def __init__(self, registry: CollectorRegistry = REGISTRY): - self.registry = registry - - def render_GET(self, request: Request) -> bytes: - request.setHeader(b"Content-Type", CONTENT_TYPE_LATEST.encode("ascii")) - response = generate_latest(self.registry) - request.setHeader(b"Content-Length", str(len(response))) - return response diff --git a/synapse/metrics/_legacy_exposition.py b/synapse/metrics/_legacy_exposition.py new file mode 100644 index 0000000000..ff640a49af --- /dev/null +++ b/synapse/metrics/_legacy_exposition.py @@ -0,0 +1,284 @@ +# Copyright 2015-2019 Prometheus Python Client Developers +# Copyright 2019 Matrix.org Foundation C.I.C. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +""" +This code is based off `prometheus_client/exposition.py` from version 0.7.1. + +Due to the renaming of metrics in prometheus_client 0.4.0, this customised +vendoring of the code will emit both the old versions that Synapse dashboards +expect, and the newer "best practice" version of the up-to-date official client. +""" + +import math +import threading +from http.server import BaseHTTPRequestHandler, HTTPServer +from socketserver import ThreadingMixIn +from typing import Any, Dict, List, Type, Union +from urllib.parse import parse_qs, urlparse + +from prometheus_client import REGISTRY, CollectorRegistry +from prometheus_client.core import Sample + +from twisted.web.resource import Resource +from twisted.web.server import Request + +from synapse.util import caches + +CONTENT_TYPE_LATEST = "text/plain; version=0.0.4; charset=utf-8" + + +def floatToGoString(d: Union[int, float]) -> str: + d = float(d) + if d == math.inf: + return "+Inf" + elif d == -math.inf: + return "-Inf" + elif math.isnan(d): + return "NaN" + else: + s = repr(d) + dot = s.find(".") + # Go switches to exponents sooner than Python. + # We only need to care about positive values for le/quantile. + if d > 0 and dot > 6: + mantissa = f"{s[0]}.{s[1:dot]}{s[dot + 1 :]}".rstrip("0.") + return f"{mantissa}e+0{dot - 1}" + return s + + +def sample_line(line: Sample, name: str) -> str: + if line.labels: + labelstr = "{{{0}}}".format( + ",".join( + [ + '{}="{}"'.format( + k, + v.replace("\\", r"\\").replace("\n", r"\n").replace('"', r"\""), + ) + for k, v in sorted(line.labels.items()) + ] + ) + ) + else: + labelstr = "" + timestamp = "" + if line.timestamp is not None: + # Convert to milliseconds. + timestamp = f" {int(float(line.timestamp) * 1000):d}" + return "{}{} {}{}\n".format(name, labelstr, floatToGoString(line.value), timestamp) + + +# Mapping from new metric names to legacy metric names. +# We translate these back to their old names when exposing them through our +# legacy vendored exporter. +# Only this legacy exposition module applies these name changes. +LEGACY_METRIC_NAMES = { + "synapse_util_caches_cache_hits": "synapse_util_caches_cache:hits", + "synapse_util_caches_cache_size": "synapse_util_caches_cache:size", + "synapse_util_caches_cache_evicted_size": "synapse_util_caches_cache:evicted_size", + "synapse_util_caches_cache_total": "synapse_util_caches_cache:total", + "synapse_util_caches_response_cache_size": "synapse_util_caches_response_cache:size", + "synapse_util_caches_response_cache_hits": "synapse_util_caches_response_cache:hits", + "synapse_util_caches_response_cache_evicted_size": "synapse_util_caches_response_cache:evicted_size", + "synapse_util_caches_response_cache_total": "synapse_util_caches_response_cache:total", +} + + +def generate_latest(registry: CollectorRegistry, emit_help: bool = False) -> bytes: + """ + Generate metrics in legacy format. Modern metrics are generated directly + by prometheus-client. + """ + + # Trigger the cache metrics to be rescraped, which updates the common + # metrics but do not produce metrics themselves + for collector in caches.collectors_by_name.values(): + collector.collect() + + output = [] + + for metric in registry.collect(): + if not metric.samples: + # No samples, don't bother. + continue + + # Translate to legacy metric name if it has one. + mname = LEGACY_METRIC_NAMES.get(metric.name, metric.name) + mnewname = metric.name + mtype = metric.type + + # OpenMetrics -> Prometheus + if mtype == "counter": + mnewname = mnewname + "_total" + elif mtype == "info": + mtype = "gauge" + mnewname = mnewname + "_info" + elif mtype == "stateset": + mtype = "gauge" + elif mtype == "gaugehistogram": + mtype = "histogram" + elif mtype == "unknown": + mtype = "untyped" + + # Output in the old format for compatibility. + if emit_help: + output.append( + "# HELP {} {}\n".format( + mname, + metric.documentation.replace("\\", r"\\").replace("\n", r"\n"), + ) + ) + output.append(f"# TYPE {mname} {mtype}\n") + + om_samples: Dict[str, List[str]] = {} + for s in metric.samples: + for suffix in ["_created", "_gsum", "_gcount"]: + if s.name == mname + suffix: + # OpenMetrics specific sample, put in a gauge at the end. + # (these come from gaugehistograms which don't get renamed, + # so no need to faff with mnewname) + om_samples.setdefault(suffix, []).append(sample_line(s, s.name)) + break + else: + newname = s.name.replace(mnewname, mname) + if ":" in newname and newname.endswith("_total"): + newname = newname[: -len("_total")] + output.append(sample_line(s, newname)) + + for suffix, lines in sorted(om_samples.items()): + if emit_help: + output.append( + "# HELP {}{} {}\n".format( + mname, + suffix, + metric.documentation.replace("\\", r"\\").replace("\n", r"\n"), + ) + ) + output.append(f"# TYPE {mname}{suffix} gauge\n") + output.extend(lines) + + # Get rid of the weird colon things while we're at it + if mtype == "counter": + mnewname = mnewname.replace(":total", "") + mnewname = mnewname.replace(":", "_") + + if mname == mnewname: + continue + + # Also output in the new format, if it's different. + if emit_help: + output.append( + "# HELP {} {}\n".format( + mnewname, + metric.documentation.replace("\\", r"\\").replace("\n", r"\n"), + ) + ) + output.append(f"# TYPE {mnewname} {mtype}\n") + + for s in metric.samples: + # Get rid of the OpenMetrics specific samples (we should already have + # dealt with them above anyway.) + for suffix in ["_created", "_gsum", "_gcount"]: + if s.name == mname + suffix: + break + else: + sample_name = LEGACY_METRIC_NAMES.get(s.name, s.name) + output.append( + sample_line(s, sample_name.replace(":total", "").replace(":", "_")) + ) + + return "".join(output).encode("utf-8") + + +class MetricsHandler(BaseHTTPRequestHandler): + """HTTP handler that gives metrics from ``REGISTRY``.""" + + registry = REGISTRY + + def do_GET(self) -> None: + registry = self.registry + params = parse_qs(urlparse(self.path).query) + + if "help" in params: + emit_help = True + else: + emit_help = False + + try: + output = generate_latest(registry, emit_help=emit_help) + except Exception: + self.send_error(500, "error generating metric output") + raise + self.send_response(200) + self.send_header("Content-Type", CONTENT_TYPE_LATEST) + self.send_header("Content-Length", str(len(output))) + self.end_headers() + self.wfile.write(output) + + def log_message(self, format: str, *args: Any) -> None: + """Log nothing.""" + + @classmethod + def factory(cls, registry: CollectorRegistry) -> Type: + """Returns a dynamic MetricsHandler class tied + to the passed registry. + """ + # This implementation relies on MetricsHandler.registry + # (defined above and defaulted to REGISTRY). + + # As we have unicode_literals, we need to create a str() + # object for type(). + cls_name = str(cls.__name__) + MyMetricsHandler = type(cls_name, (cls, object), {"registry": registry}) + return MyMetricsHandler + + +class _ThreadingSimpleServer(ThreadingMixIn, HTTPServer): + """Thread per request HTTP server.""" + + # Make worker threads "fire and forget". Beginning with Python 3.7 this + # prevents a memory leak because ``ThreadingMixIn`` starts to gather all + # non-daemon threads in a list in order to join on them at server close. + # Enabling daemon threads virtually makes ``_ThreadingSimpleServer`` the + # same as Python 3.7's ``ThreadingHTTPServer``. + daemon_threads = True + + +def start_http_server( + port: int, addr: str = "", registry: CollectorRegistry = REGISTRY +) -> None: + """Starts an HTTP server for prometheus metrics as a daemon thread""" + CustomMetricsHandler = MetricsHandler.factory(registry) + httpd = _ThreadingSimpleServer((addr, port), CustomMetricsHandler) + t = threading.Thread(target=httpd.serve_forever) + t.daemon = True + t.start() + + +class MetricsResource(Resource): + """ + Twisted ``Resource`` that serves prometheus metrics. + """ + + isLeaf = True + + def __init__(self, registry: CollectorRegistry = REGISTRY): + self.registry = registry + + def render_GET(self, request: Request) -> bytes: + request.setHeader(b"Content-Type", CONTENT_TYPE_LATEST.encode("ascii")) + response = generate_latest(self.registry) + request.setHeader(b"Content-Length", str(len(response))) + return response diff --git a/synapse/util/caches/__init__.py b/synapse/util/caches/__init__.py index 42f6abb5e1..bdf9b0dc8c 100644 --- a/synapse/util/caches/__init__.py +++ b/synapse/util/caches/__init__.py @@ -34,10 +34,10 @@ TRACK_MEMORY_USAGE = False caches_by_name: Dict[str, Sized] = {} collectors_by_name: Dict[str, "CacheMetric"] = {} -cache_size = Gauge("synapse_util_caches_cache:size", "", ["name"]) -cache_hits = Gauge("synapse_util_caches_cache:hits", "", ["name"]) -cache_evicted = Gauge("synapse_util_caches_cache:evicted_size", "", ["name", "reason"]) -cache_total = Gauge("synapse_util_caches_cache:total", "", ["name"]) +cache_size = Gauge("synapse_util_caches_cache_size", "", ["name"]) +cache_hits = Gauge("synapse_util_caches_cache_hits", "", ["name"]) +cache_evicted = Gauge("synapse_util_caches_cache_evicted_size", "", ["name", "reason"]) +cache_total = Gauge("synapse_util_caches_cache_total", "", ["name"]) cache_max_size = Gauge("synapse_util_caches_cache_max_size", "", ["name"]) cache_memory_usage = Gauge( "synapse_util_caches_cache_size_bytes", @@ -45,12 +45,12 @@ cache_memory_usage = Gauge( ["name"], ) -response_cache_size = Gauge("synapse_util_caches_response_cache:size", "", ["name"]) -response_cache_hits = Gauge("synapse_util_caches_response_cache:hits", "", ["name"]) +response_cache_size = Gauge("synapse_util_caches_response_cache_size", "", ["name"]) +response_cache_hits = Gauge("synapse_util_caches_response_cache_hits", "", ["name"]) response_cache_evicted = Gauge( - "synapse_util_caches_response_cache:evicted_size", "", ["name", "reason"] + "synapse_util_caches_response_cache_evicted_size", "", ["name", "reason"] ) -response_cache_total = Gauge("synapse_util_caches_response_cache:total", "", ["name"]) +response_cache_total = Gauge("synapse_util_caches_response_cache_total", "", ["name"]) class EvictionReason(Enum): diff --git a/tests/test_metrics.py b/tests/test_metrics.py index b4574b2ffe..1a70eddc9b 100644 --- a/tests/test_metrics.py +++ b/tests/test_metrics.py @@ -12,7 +12,16 @@ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. +try: + from importlib import metadata +except ImportError: + import importlib_metadata as metadata # type: ignore[no-redef] +from unittest.mock import patch + +from pkg_resources import parse_version + +from synapse.app._base import _set_prometheus_client_use_created_metrics from synapse.metrics import REGISTRY, InFlightGauge, generate_latest from synapse.util.caches.deferred_cache import DeferredCache @@ -162,3 +171,30 @@ class CacheMetricsTests(unittest.HomeserverTestCase): self.assertEqual(items["synapse_util_caches_cache_size"], "1.0") self.assertEqual(items["synapse_util_caches_cache_max_size"], "777.0") + + +class PrometheusMetricsHackTestCase(unittest.HomeserverTestCase): + if parse_version(metadata.version("prometheus_client")) < parse_version("0.14.0"): + skip = "prometheus-client too old" + + def test_created_metrics_disabled(self) -> None: + """ + Tests that a brittle hack, to disable `_created` metrics, works. + This involves poking at the internals of prometheus-client. + It's not the end of the world if this doesn't work. + + This test gives us a way to notice if prometheus-client changes + their internals. + """ + import prometheus_client.metrics + + PRIVATE_FLAG_NAME = "_use_created" + + # By default, the pesky `_created` metrics are enabled. + # Check this assumption is still valid. + self.assertTrue(getattr(prometheus_client.metrics, PRIVATE_FLAG_NAME)) + + with patch("prometheus_client.metrics") as mock: + setattr(mock, PRIVATE_FLAG_NAME, True) + _set_prometheus_client_use_created_metrics(False) + self.assertFalse(getattr(mock, PRIVATE_FLAG_NAME, False)) -- cgit 1.5.1 From 1eea73b4133da3afc2361592f7b2fa4a4125249d Mon Sep 17 00:00:00 2001 From: Eric Eastwood Date: Tue, 30 Aug 2022 06:08:29 -0500 Subject: Fix rate limit metrics registering twice and misreporting (#13649) * Fix rate limit metrics registering twice and misreporting Fix https://github.com/matrix-org/synapse/issues/13641 * Fix lints * Add changelog * Document `metrics_name=None`. --- changelog.d/13649.bugfix | 1 + synapse/server.py | 4 +- synapse/util/ratelimitutils.py | 155 ++++++++++++++++++++++++++++++++--------- 3 files changed, 128 insertions(+), 32 deletions(-) create mode 100644 changelog.d/13649.bugfix (limited to 'synapse/util') diff --git a/changelog.d/13649.bugfix b/changelog.d/13649.bugfix new file mode 100644 index 0000000000..e6513a585a --- /dev/null +++ b/changelog.d/13649.bugfix @@ -0,0 +1 @@ +Fix rate limit gauge metrics registering twice and misreporting (`synapse_rate_limit_sleep_affected_hosts`, `synapse_rate_limit_reject_affected_hosts`). diff --git a/synapse/server.py b/synapse/server.py index 181984a1a4..c2e55bf0b1 100644 --- a/synapse/server.py +++ b/synapse/server.py @@ -756,7 +756,9 @@ class HomeServer(metaclass=abc.ABCMeta): @cache_in_self def get_federation_ratelimiter(self) -> FederationRateLimiter: return FederationRateLimiter( - self.get_clock(), config=self.config.ratelimiting.rc_federation + self.get_clock(), + config=self.config.ratelimiting.rc_federation, + metrics_name="federation_servlets", ) @cache_in_self diff --git a/synapse/util/ratelimitutils.py b/synapse/util/ratelimitutils.py index f678b52cb4..9f64fed0d7 100644 --- a/synapse/util/ratelimitutils.py +++ b/synapse/util/ratelimitutils.py @@ -15,10 +15,23 @@ import collections import contextlib import logging +import threading import typing -from typing import Any, DefaultDict, Iterator, List, Set +from typing import ( + Any, + Callable, + DefaultDict, + Dict, + Iterator, + List, + Mapping, + Optional, + Set, + Tuple, +) from prometheus_client.core import Counter +from typing_extensions import ContextManager from twisted.internet import defer @@ -40,12 +53,20 @@ logger = logging.getLogger(__name__) # Track how much the ratelimiter is affecting requests -rate_limit_sleep_counter = Counter("synapse_rate_limit_sleep", "") -rate_limit_reject_counter = Counter("synapse_rate_limit_reject", "") +rate_limit_sleep_counter = Counter( + "synapse_rate_limit_sleep", + "Number of requests slept by the rate limiter", + ["rate_limiter_name"], +) +rate_limit_reject_counter = Counter( + "synapse_rate_limit_reject", + "Number of requests rejected by the rate limiter", + ["rate_limiter_name"], +) queue_wait_timer = Histogram( "synapse_rate_limit_queue_wait_time_seconds", - "sec", - [], + "Amount of time spent waiting for the rate limiter to let our request through.", + ["rate_limiter_name"], buckets=( 0.005, 0.01, @@ -65,35 +86,92 @@ queue_wait_timer = Histogram( ) +_rate_limiter_instances: Set["FederationRateLimiter"] = set() +# Protects the _rate_limiter_instances set from concurrent access +_rate_limiter_instances_lock = threading.Lock() + + +def _get_counts_from_rate_limiter_instance( + count_func: Callable[["FederationRateLimiter"], int] +) -> Mapping[Tuple[str, ...], int]: + """Returns a count of something (slept/rejected hosts) by (metrics_name)""" + # Cast to a list to prevent it changing while the Prometheus + # thread is collecting metrics + with _rate_limiter_instances_lock: + rate_limiter_instances = list(_rate_limiter_instances) + + # Map from (metrics_name,) -> int, the number of something like slept hosts + # or rejected hosts. The key type is Tuple[str], but we leave the length + # unspecified for compatability with LaterGauge's annotations. + counts: Dict[Tuple[str, ...], int] = {} + for rate_limiter_instance in rate_limiter_instances: + # Only track metrics if they provided a `metrics_name` to + # differentiate this instance of the rate limiter. + if rate_limiter_instance.metrics_name: + key = (rate_limiter_instance.metrics_name,) + counts[key] = count_func(rate_limiter_instance) + + return counts + + +# We track the number of affected hosts per time-period so we can +# differentiate one really noisy homeserver from a general +# ratelimit tuning problem across the federation. +LaterGauge( + "synapse_rate_limit_sleep_affected_hosts", + "Number of hosts that had requests put to sleep", + ["rate_limiter_name"], + lambda: _get_counts_from_rate_limiter_instance( + lambda rate_limiter_instance: sum( + ratelimiter.should_sleep() + for ratelimiter in rate_limiter_instance.ratelimiters.values() + ) + ), +) +LaterGauge( + "synapse_rate_limit_reject_affected_hosts", + "Number of hosts that had requests rejected", + ["rate_limiter_name"], + lambda: _get_counts_from_rate_limiter_instance( + lambda rate_limiter_instance: sum( + ratelimiter.should_reject() + for ratelimiter in rate_limiter_instance.ratelimiters.values() + ) + ), +) + + class FederationRateLimiter: - def __init__(self, clock: Clock, config: FederationRatelimitSettings): + """Used to rate limit request per-host.""" + + def __init__( + self, + clock: Clock, + config: FederationRatelimitSettings, + metrics_name: Optional[str] = None, + ): + """ + Args: + clock + config + metrics_name: The name of the rate limiter so we can differentiate it + from the rest in the metrics. If `None`, we don't track metrics + for this rate limiter. + + """ + self.metrics_name = metrics_name + def new_limiter() -> "_PerHostRatelimiter": - return _PerHostRatelimiter(clock=clock, config=config) + return _PerHostRatelimiter( + clock=clock, config=config, metrics_name=metrics_name + ) self.ratelimiters: DefaultDict[ str, "_PerHostRatelimiter" ] = collections.defaultdict(new_limiter) - # We track the number of affected hosts per time-period so we can - # differentiate one really noisy homeserver from a general - # ratelimit tuning problem across the federation. - LaterGauge( - "synapse_rate_limit_sleep_affected_hosts", - "Number of hosts that had requests put to sleep", - [], - lambda: sum( - ratelimiter.should_sleep() for ratelimiter in self.ratelimiters.values() - ), - ) - LaterGauge( - "synapse_rate_limit_reject_affected_hosts", - "Number of hosts that had requests rejected", - [], - lambda: sum( - ratelimiter.should_reject() - for ratelimiter in self.ratelimiters.values() - ), - ) + with _rate_limiter_instances_lock: + _rate_limiter_instances.add(self) def ratelimit(self, host: str) -> "_GeneratorContextManager[defer.Deferred[None]]": """Used to ratelimit an incoming request from a given host @@ -114,13 +192,23 @@ class FederationRateLimiter: class _PerHostRatelimiter: - def __init__(self, clock: Clock, config: FederationRatelimitSettings): + def __init__( + self, + clock: Clock, + config: FederationRatelimitSettings, + metrics_name: Optional[str] = None, + ): """ Args: clock config + metrics_name: The name of the rate limiter so we can differentiate it + from the rest in the metrics. If `None`, we don't track metrics + for this rate limiter. + from the rest in the metrics """ self.clock = clock + self.metrics_name = metrics_name self.window_size = config.window_size self.sleep_limit = config.sleep_limit @@ -178,7 +266,10 @@ class _PerHostRatelimiter: return len(self.request_times) > self.sleep_limit async def _on_enter_with_tracing(self, request_id: object) -> None: - with start_active_span("ratelimit wait"), queue_wait_timer.time(): + maybe_metrics_cm: ContextManager = contextlib.nullcontext() + if self.metrics_name: + maybe_metrics_cm = queue_wait_timer.labels(self.metrics_name).time() + with start_active_span("ratelimit wait"), maybe_metrics_cm: await self._on_enter(request_id) def _on_enter(self, request_id: object) -> "defer.Deferred[None]": @@ -193,7 +284,8 @@ class _PerHostRatelimiter: # sleeping or in the ready queue). if self.should_reject(): logger.debug("Ratelimiter(%s): rejecting request", self.host) - rate_limit_reject_counter.inc() + if self.metrics_name: + rate_limit_reject_counter.labels(self.metrics_name).inc() raise LimitExceededError( retry_after_ms=int(self.window_size / self.sleep_limit) ) @@ -228,7 +320,8 @@ class _PerHostRatelimiter: id(request_id), self.sleep_sec, ) - rate_limit_sleep_counter.inc() + if self.metrics_name: + rate_limit_sleep_counter.labels(self.metrics_name).inc() ret_defer = run_in_background(self.clock.sleep, self.sleep_sec) self.sleeping_requests.add(request_id) -- cgit 1.5.1 From 7bc110a19e6de0572b0c9513726d13298b45ced2 Mon Sep 17 00:00:00 2001 From: reivilibre Date: Wed, 31 Aug 2022 11:16:05 +0000 Subject: Generalise the `@cancellable` annotation so it can be used on functions other than just servlet methods. (#13662) --- changelog.d/13662.misc | 1 + synapse/federation/transport/server/_base.py | 5 +- synapse/http/server.py | 68 ++----------------------- synapse/replication/http/_base.py | 7 +-- synapse/rest/client/room.py | 3 +- synapse/util/cancellation.py | 56 ++++++++++++++++++++ tests/federation/transport/server/test__base.py | 3 +- tests/http/test_servlet.py | 2 +- tests/replication/http/test__base.py | 3 +- tests/test_server.py | 2 +- 10 files changed, 75 insertions(+), 75 deletions(-) create mode 100644 changelog.d/13662.misc create mode 100644 synapse/util/cancellation.py (limited to 'synapse/util') diff --git a/changelog.d/13662.misc b/changelog.d/13662.misc new file mode 100644 index 0000000000..3dea4a1c2c --- /dev/null +++ b/changelog.d/13662.misc @@ -0,0 +1 @@ +Generalise the `@cancellable` annotation so it can be used on functions other than just servlet methods. \ No newline at end of file diff --git a/synapse/federation/transport/server/_base.py b/synapse/federation/transport/server/_base.py index bb0f8d6b7b..1db8009d6c 100644 --- a/synapse/federation/transport/server/_base.py +++ b/synapse/federation/transport/server/_base.py @@ -21,7 +21,7 @@ from typing import TYPE_CHECKING, Any, Awaitable, Callable, Dict, Optional, Tupl from synapse.api.errors import Codes, FederationDeniedError, SynapseError from synapse.api.urls import FEDERATION_V1_PREFIX -from synapse.http.server import HttpServer, ServletCallback, is_method_cancellable +from synapse.http.server import HttpServer, ServletCallback from synapse.http.servlet import parse_json_object_from_request from synapse.http.site import SynapseRequest from synapse.logging.context import run_in_background @@ -34,6 +34,7 @@ from synapse.logging.opentracing import ( whitelisted_homeserver, ) from synapse.types import JsonDict +from synapse.util.cancellation import is_function_cancellable from synapse.util.ratelimitutils import FederationRateLimiter from synapse.util.stringutils import parse_and_validate_server_name @@ -375,7 +376,7 @@ class BaseFederationServlet: if code is None: continue - if is_method_cancellable(code): + if is_function_cancellable(code): # The wrapper added by `self._wrap` will inherit the cancellable flag, # but the wrapper itself does not support cancellation yet. # Once resolved, the cancellation tests in diff --git a/synapse/http/server.py b/synapse/http/server.py index 19f42159b8..6068a94b40 100644 --- a/synapse/http/server.py +++ b/synapse/http/server.py @@ -33,7 +33,6 @@ from typing import ( Optional, Pattern, Tuple, - TypeVar, Union, ) @@ -64,6 +63,7 @@ from synapse.logging.context import defer_to_thread, preserve_fn, run_in_backgro from synapse.logging.opentracing import active_span, start_active_span, trace_servlet from synapse.util import json_encoder from synapse.util.caches import intern_dict +from synapse.util.cancellation import is_function_cancellable from synapse.util.iterutils import chunk_seq if TYPE_CHECKING: @@ -94,68 +94,6 @@ HTML_ERROR_TEMPLATE = """ HTTP_STATUS_REQUEST_CANCELLED = 499 -F = TypeVar("F", bound=Callable[..., Any]) - - -_cancellable_method_names = frozenset( - { - # `RestServlet`, `BaseFederationServlet` and `BaseFederationServerServlet` - # methods - "on_GET", - "on_PUT", - "on_POST", - "on_DELETE", - # `_AsyncResource`, `DirectServeHtmlResource` and `DirectServeJsonResource` - # methods - "_async_render_GET", - "_async_render_PUT", - "_async_render_POST", - "_async_render_DELETE", - "_async_render_OPTIONS", - # `ReplicationEndpoint` methods - "_handle_request", - } -) - - -def cancellable(method: F) -> F: - """Marks a servlet method as cancellable. - - Methods with this decorator will be cancelled if the client disconnects before we - finish processing the request. - - During cancellation, `Deferred.cancel()` will be invoked on the `Deferred` wrapping - the method. The `cancel()` call will propagate down to the `Deferred` that is - currently being waited on. That `Deferred` will raise a `CancelledError`, which will - propagate up, as per normal exception handling. - - Before applying this decorator to a new endpoint, you MUST recursively check - that all `await`s in the function are on `async` functions or `Deferred`s that - handle cancellation cleanly, otherwise a variety of bugs may occur, ranging from - premature logging context closure, to stuck requests, to database corruption. - - Usage: - class SomeServlet(RestServlet): - @cancellable - async def on_GET(self, request: SynapseRequest) -> ...: - ... - """ - if method.__name__ not in _cancellable_method_names and not any( - method.__name__.startswith(prefix) for prefix in _cancellable_method_names - ): - raise ValueError( - "@cancellable decorator can only be applied to servlet methods." - ) - - method.cancellable = True # type: ignore[attr-defined] - return method - - -def is_method_cancellable(method: Callable[..., Any]) -> bool: - """Checks whether a servlet method has the `@cancellable` flag.""" - return getattr(method, "cancellable", False) - - def return_json_error( f: failure.Failure, request: SynapseRequest, config: Optional[HomeServerConfig] ) -> None: @@ -389,7 +327,7 @@ class _AsyncResource(resource.Resource, metaclass=abc.ABCMeta): method_handler = getattr(self, "_async_render_%s" % (request_method,), None) if method_handler: - request.is_render_cancellable = is_method_cancellable(method_handler) + request.is_render_cancellable = is_function_cancellable(method_handler) raw_callback_return = method_handler(request) @@ -551,7 +489,7 @@ class JsonResource(DirectServeJsonResource): async def _async_render(self, request: SynapseRequest) -> Tuple[int, Any]: callback, servlet_classname, group_dict = self._get_handler_for_request(request) - request.is_render_cancellable = is_method_cancellable(callback) + request.is_render_cancellable = is_function_cancellable(callback) # Make sure we have an appropriate name for this handler in prometheus # (rather than the default of JsonResource). diff --git a/synapse/replication/http/_base.py b/synapse/replication/http/_base.py index 561ad5bf04..acb0bd18f7 100644 --- a/synapse/replication/http/_base.py +++ b/synapse/replication/http/_base.py @@ -26,12 +26,13 @@ from twisted.web.server import Request from synapse.api.errors import HttpResponseException, SynapseError from synapse.http import RequestTimedOutError -from synapse.http.server import HttpServer, is_method_cancellable +from synapse.http.server import HttpServer from synapse.http.site import SynapseRequest from synapse.logging import opentracing from synapse.logging.opentracing import trace_with_opname from synapse.types import JsonDict from synapse.util.caches.response_cache import ResponseCache +from synapse.util.cancellation import is_function_cancellable from synapse.util.stringutils import random_string if TYPE_CHECKING: @@ -311,7 +312,7 @@ class ReplicationEndpoint(metaclass=abc.ABCMeta): url_args = list(self.PATH_ARGS) method = self.METHOD - if self.CACHE and is_method_cancellable(self._handle_request): + if self.CACHE and is_function_cancellable(self._handle_request): raise Exception( f"{self.__class__.__name__} has been marked as cancellable, but CACHE " "is set. The cancellable flag would have no effect." @@ -359,6 +360,6 @@ class ReplicationEndpoint(metaclass=abc.ABCMeta): # The `@cancellable` decorator may be applied to `_handle_request`. But we # told `HttpServer.register_paths` that our handler is `_check_auth_and_handle`, # so we have to set up the cancellable flag ourselves. - request.is_render_cancellable = is_method_cancellable(self._handle_request) + request.is_render_cancellable = is_function_cancellable(self._handle_request) return await self._handle_request(request, **kwargs) diff --git a/synapse/rest/client/room.py b/synapse/rest/client/room.py index 3259de4802..0e2834008e 100644 --- a/synapse/rest/client/room.py +++ b/synapse/rest/client/room.py @@ -37,7 +37,7 @@ from synapse.api.errors import ( ) from synapse.api.filtering import Filter from synapse.events.utils import format_event_for_client_v2 -from synapse.http.server import HttpServer, cancellable +from synapse.http.server import HttpServer from synapse.http.servlet import ( ResolveRoomIdMixin, RestServlet, @@ -57,6 +57,7 @@ from synapse.storage.state import StateFilter from synapse.streams.config import PaginationConfig from synapse.types import JsonDict, StreamToken, ThirdPartyInstanceID, UserID from synapse.util import json_decoder +from synapse.util.cancellation import cancellable from synapse.util.stringutils import parse_and_validate_server_name, random_string if TYPE_CHECKING: diff --git a/synapse/util/cancellation.py b/synapse/util/cancellation.py new file mode 100644 index 0000000000..472d2e3aeb --- /dev/null +++ b/synapse/util/cancellation.py @@ -0,0 +1,56 @@ +# Copyright 2022 The Matrix.org Foundation C.I.C. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. +from typing import Any, Callable, TypeVar + +F = TypeVar("F", bound=Callable[..., Any]) + + +def cancellable(function: F) -> F: + """Marks a function as cancellable. + + Servlet methods with this decorator will be cancelled if the client disconnects before we + finish processing the request. + + Although this annotation is particularly useful for servlet methods, it's also + useful for intermediate functions, where it documents the fact that the function has + been audited for cancellation safety and needs to preserve that. + This then simplifies auditing new functions that call those same intermediate + functions. + + During cancellation, `Deferred.cancel()` will be invoked on the `Deferred` wrapping + the method. The `cancel()` call will propagate down to the `Deferred` that is + currently being waited on. That `Deferred` will raise a `CancelledError`, which will + propagate up, as per normal exception handling. + + Before applying this decorator to a new function, you MUST recursively check + that all `await`s in the function are on `async` functions or `Deferred`s that + handle cancellation cleanly, otherwise a variety of bugs may occur, ranging from + premature logging context closure, to stuck requests, to database corruption. + + See the documentation page on Cancellation for more information. + + Usage: + class SomeServlet(RestServlet): + @cancellable + async def on_GET(self, request: SynapseRequest) -> ...: + ... + """ + + function.cancellable = True # type: ignore[attr-defined] + return function + + +def is_function_cancellable(function: Callable[..., Any]) -> bool: + """Checks whether a servlet method has the `@cancellable` flag.""" + return getattr(function, "cancellable", False) diff --git a/tests/federation/transport/server/test__base.py b/tests/federation/transport/server/test__base.py index d33e86db4c..e88e5d8bb3 100644 --- a/tests/federation/transport/server/test__base.py +++ b/tests/federation/transport/server/test__base.py @@ -18,9 +18,10 @@ from typing import Dict, List, Tuple from synapse.api.errors import Codes from synapse.federation.transport.server import BaseFederationServlet from synapse.federation.transport.server._base import Authenticator, _parse_auth_header -from synapse.http.server import JsonResource, cancellable +from synapse.http.server import JsonResource from synapse.server import HomeServer from synapse.types import JsonDict +from synapse.util.cancellation import cancellable from synapse.util.ratelimitutils import FederationRateLimiter from tests import unittest diff --git a/tests/http/test_servlet.py b/tests/http/test_servlet.py index bb966c80c6..3cbca0f5a3 100644 --- a/tests/http/test_servlet.py +++ b/tests/http/test_servlet.py @@ -18,7 +18,6 @@ from typing import Tuple from unittest.mock import Mock from synapse.api.errors import Codes, SynapseError -from synapse.http.server import cancellable from synapse.http.servlet import ( RestServlet, parse_json_object_from_request, @@ -28,6 +27,7 @@ from synapse.http.site import SynapseRequest from synapse.rest.client._base import client_patterns from synapse.server import HomeServer from synapse.types import JsonDict +from synapse.util.cancellation import cancellable from tests import unittest from tests.http.server._base import test_disconnect diff --git a/tests/replication/http/test__base.py b/tests/replication/http/test__base.py index 822a957c3a..936ab4504a 100644 --- a/tests/replication/http/test__base.py +++ b/tests/replication/http/test__base.py @@ -18,11 +18,12 @@ from typing import Tuple from twisted.web.server import Request from synapse.api.errors import Codes -from synapse.http.server import JsonResource, cancellable +from synapse.http.server import JsonResource from synapse.replication.http import REPLICATION_PREFIX from synapse.replication.http._base import ReplicationEndpoint from synapse.server import HomeServer from synapse.types import JsonDict +from synapse.util.cancellation import cancellable from tests import unittest from tests.http.server._base import test_disconnect diff --git a/tests/test_server.py b/tests/test_server.py index d2b2d8344a..23975d59c3 100644 --- a/tests/test_server.py +++ b/tests/test_server.py @@ -26,12 +26,12 @@ from synapse.http.server import ( DirectServeJsonResource, JsonResource, OptionsResource, - cancellable, ) from synapse.http.site import SynapseRequest, SynapseSite from synapse.logging.context import make_deferred_yieldable from synapse.types import JsonDict from synapse.util import Clock +from synapse.util.cancellation import cancellable from tests import unittest from tests.http.server._base import test_disconnect -- cgit 1.5.1 From b455c2a5ec4874a6897c0448631d8ab8f36f9339 Mon Sep 17 00:00:00 2001 From: reivilibre Date: Tue, 6 Sep 2022 11:21:21 +0000 Subject: Update Grafana dashboard to not use legacy metric names. (#13714) --- changelog.d/13714.misc | 1 + contrib/grafana/synapse.json | 138 +++++++++++++++++----------------- synapse/metrics/_legacy_exposition.py | 4 +- synapse/util/caches/__init__.py | 4 +- 4 files changed, 74 insertions(+), 73 deletions(-) create mode 100644 changelog.d/13714.misc (limited to 'synapse/util') diff --git a/changelog.d/13714.misc b/changelog.d/13714.misc new file mode 100644 index 0000000000..07ace50b12 --- /dev/null +++ b/changelog.d/13714.misc @@ -0,0 +1 @@ +Add experimental configuration option to allow disabling legacy Prometheus metric names. \ No newline at end of file diff --git a/contrib/grafana/synapse.json b/contrib/grafana/synapse.json index 248cd6d9ad..58061e2fce 100644 --- a/contrib/grafana/synapse.json +++ b/contrib/grafana/synapse.json @@ -335,7 +335,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "sum(rate(synapse_storage_events_persisted_events{instance=\"$instance\"}[$bucket_size]))", + "expr": "sum(rate(synapse_storage_events_persisted_events_total{instance=\"$instance\"}[$bucket_size]))", "hide": false, "instant": false, "legendFormat": "Events", @@ -1423,7 +1423,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "rate(synapse_background_process_ru_utime_seconds{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size])+rate(synapse_background_process_ru_stime_seconds{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size])", + "expr": "rate(synapse_background_process_ru_utime_seconds_total{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size])+rate(synapse_background_process_ru_stime_seconds_total{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size])", "format": "time_series", "hide": false, "instant": false, @@ -1804,7 +1804,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "sum(rate(synapse_storage_events_persisted_events{instance=\"$instance\"}[$bucket_size])) without (job,index)", + "expr": "sum(rate(synapse_storage_events_persisted_events_total{instance=\"$instance\"}[$bucket_size])) without (job,index)", "format": "time_series", "interval": "", "intervalFactor": 2, @@ -2437,7 +2437,7 @@ "uid": "$datasource" }, "exemplar": false, - "expr": "sum(rate(synapse_state_res_db_for_biggest_room_seconds{instance=\"$instance\"}[1m]))", + "expr": "sum(rate(synapse_state_res_db_for_biggest_room_seconds_total{instance=\"$instance\"}[1m]))", "format": "time_series", "hide": false, "instant": false, @@ -2451,7 +2451,7 @@ "uid": "$datasource" }, "exemplar": false, - "expr": "sum(rate(synapse_state_res_cpu_for_biggest_room_seconds{instance=\"$instance\"}[1m]))", + "expr": "sum(rate(synapse_state_res_cpu_for_biggest_room_seconds_total{instance=\"$instance\"}[1m]))", "format": "time_series", "hide": false, "instant": false, @@ -3425,7 +3425,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "rate(synapse_background_process_ru_utime_seconds{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size])+rate(synapse_background_process_ru_stime_seconds{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size])", + "expr": "rate(synapse_background_process_ru_utime_seconds_total{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size])+rate(synapse_background_process_ru_stime_seconds_total{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size])", "format": "time_series", "interval": "", "intervalFactor": 1, @@ -3518,7 +3518,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "rate(synapse_background_process_db_txn_duration_seconds{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size]) + rate(synapse_background_process_db_sched_duration_seconds{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size])", + "expr": "rate(synapse_background_process_db_txn_duration_seconds_total{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size]) + rate(synapse_background_process_db_sched_duration_seconds_total{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size])", "format": "time_series", "hide": false, "intervalFactor": 1, @@ -3726,7 +3726,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "sum(rate(synapse_federation_client_sent_transactions{instance=\"$instance\"}[$bucket_size]))", + "expr": "sum(rate(synapse_federation_client_sent_transactions_total{instance=\"$instance\"}[$bucket_size]))", "format": "time_series", "intervalFactor": 1, "legendFormat": "successful txn rate", @@ -3736,7 +3736,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "sum(rate(synapse_util_metrics_block_count{block_name=\"_send_new_transaction\",instance=\"$instance\"}[$bucket_size]) - ignoring (block_name) rate(synapse_federation_client_sent_transactions{instance=\"$instance\"}[$bucket_size]))", + "expr": "sum(rate(synapse_util_metrics_block_count_total{block_name=\"_send_new_transaction\",instance=\"$instance\"}[$bucket_size]) - ignoring (block_name) rate(synapse_federation_client_sent_transactions_total{instance=\"$instance\"}[$bucket_size]))", "legendFormat": "failed txn rate", "refId": "B" } @@ -3826,7 +3826,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "sum(rate(synapse_federation_server_received_pdus{instance=~\"$instance\"}[$bucket_size]))", + "expr": "sum(rate(synapse_federation_server_received_pdus_total{instance=~\"$instance\"}[$bucket_size]))", "format": "time_series", "intervalFactor": 1, "legendFormat": "pdus", @@ -3836,7 +3836,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "sum(rate(synapse_federation_server_received_edus{instance=~\"$instance\"}[$bucket_size]))", + "expr": "sum(rate(synapse_federation_server_received_edus_total{instance=~\"$instance\"}[$bucket_size]))", "format": "time_series", "intervalFactor": 1, "legendFormat": "edus", @@ -3928,7 +3928,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "sum(rate(synapse_federation_client_sent_pdu_destinations:total{instance=\"$instance\"}[$bucket_size]))", + "expr": "sum(rate(synapse_federation_client_sent_pdu_destinations:total_total{instance=\"$instance\"}[$bucket_size]))", "format": "time_series", "interval": "", "intervalFactor": 1, @@ -3939,7 +3939,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "sum(rate(synapse_federation_client_sent_edus{instance=\"$instance\"}[$bucket_size]))", + "expr": "sum(rate(synapse_federation_client_sent_edus_total{instance=\"$instance\"}[$bucket_size]))", "format": "time_series", "intervalFactor": 1, "legendFormat": "edus", @@ -5042,7 +5042,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "rate(synapse_http_httppusher_http_pushes_processed{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size]) and on (instance, job, index) (synapse_http_httppusher_http_pushes_failed + synapse_http_httppusher_http_pushes_processed) > 0", + "expr": "rate(synapse_http_httppusher_http_pushes_processed_total{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size]) and on (instance, job, index) (synapse_http_httppusher_http_pushes_failed_total + synapse_http_httppusher_http_pushes_processed_total) > 0", "format": "time_series", "interval": "", "intervalFactor": 2, @@ -5054,7 +5054,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "rate(synapse_http_httppusher_http_pushes_failed{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size]) and on (instance, job, index) (synapse_http_httppusher_http_pushes_failed + synapse_http_httppusher_http_pushes_processed) > 0", + "expr": "rate(synapse_http_httppusher_http_pushes_failed_total{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size]) and on (instance, job, index) (synapse_http_httppusher_http_pushes_failed_total + synapse_http_httppusher_http_pushes_processed_total) > 0", "format": "time_series", "intervalFactor": 2, "legendFormat": "failed {{job}}", @@ -5268,12 +5268,12 @@ "uid": "${DS_PROMETHEUS}" }, "exemplar": true, - "expr": "sum(rate(synapse_push_bulk_push_rule_evaluator_push_rules_state_size_counter{job=\"$job\",index=~\"$index\",instance=\"$instance\"}[$bucket_size]))", + "expr": "sum(rate(synapse_push_bulk_push_rule_evaluator_push_rules_state_size_counter_total{job=\"$job\",index=~\"$index\",instance=\"$instance\"}[$bucket_size]))", "format": "time_series", "interval": "", "intervalFactor": 2, "legendFormat": "{{index}}", - "metric": "synapse_push_bulk_push_rule_evaluator_push_rules_state_size_counter", + "metric": "synapse_push_bulk_push_rule_evaluator_push_rules_state_size_counter_total", "refId": "A", "step": 2 } @@ -5369,12 +5369,12 @@ "uid": "$datasource" }, "exemplar": true, - "expr": "sum(rate(synapse_push_bulk_push_rule_evaluator_push_rules_invalidation_counter{job=\"$job\",index=~\"$index\",instance=\"$instance\"}[$bucket_size]))", + "expr": "sum(rate(synapse_push_bulk_push_rule_evaluator_push_rules_invalidation_counter_total{job=\"$job\",index=~\"$index\",instance=\"$instance\"}[$bucket_size]))", "format": "time_series", "interval": "", "intervalFactor": 2, "legendFormat": "{{index}}", - "metric": "synapse_push_bulk_push_rule_evaluator_push_rules_invalidation_counter", + "metric": "synapse_push_bulk_push_rule_evaluator_push_rules_invalidation_counter_total", "refId": "A", "step": 2 } @@ -5475,12 +5475,12 @@ "uid": "${DS_PROMETHEUS}" }, "exemplar": true, - "expr": "sum(rate(synapse_util_caches_cache:hits{job=\"$job\",index=~\"$index\",name=\"push_rules_delta_state_cache_metric\",instance=\"$instance\"}[$bucket_size]))/sum(rate(synapse_util_caches_cache:total{job=\"$job\",index=~\"$index\", name=\"push_rules_delta_state_cache_metric\",instance=\"$instance\"}[$bucket_size]))", + "expr": "sum(rate(synapse_util_caches_cache_hits{job=\"$job\",index=~\"$index\",name=\"push_rules_delta_state_cache_metric\",instance=\"$instance\"}[$bucket_size]))/sum(rate(synapse_util_caches_cache{job=\"$job\",index=~\"$index\", name=\"push_rules_delta_state_cache_metric\",instance=\"$instance\"}[$bucket_size]))", "format": "time_series", "interval": "", "intervalFactor": 2, "legendFormat": "Hit Rate", - "metric": "synapse_push_bulk_push_rule_evaluator_push_rules_invalidation_counter", + "metric": "synapse_push_bulk_push_rule_evaluator_push_rules_invalidation_counter_total", "refId": "A", "step": 2 }, @@ -5490,7 +5490,7 @@ "uid": "${DS_PROMETHEUS}" }, "exemplar": true, - "expr": "sum(rate(synapse_util_caches_cache:total{job=\"$job\",index=~\"$index\", name=\"push_rules_delta_state_cache_metric\",instance=\"$instance\"}[$bucket_size]))", + "expr": "sum(rate(synapse_util_caches_cache{job=\"$job\",index=~\"$index\", name=\"push_rules_delta_state_cache_metric\",instance=\"$instance\"}[$bucket_size]))", "format": "time_series", "interval": "", "intervalFactor": 2, @@ -5598,12 +5598,12 @@ "uid": "${DS_PROMETHEUS}" }, "exemplar": true, - "expr": "sum(rate(synapse_util_caches_cache:hits{job=\"$job\",index=~\"$index\",name=\"room_push_rule_cache\",instance=\"$instance\"}[$bucket_size]))/sum(rate(synapse_util_caches_cache:total{job=\"$job\",index=~\"$index\", name=\"room_push_rule_cache\",instance=\"$instance\"}[$bucket_size]))", + "expr": "sum(rate(synapse_util_caches_cache_hits{job=\"$job\",index=~\"$index\",name=\"room_push_rule_cache\",instance=\"$instance\"}[$bucket_size]))/sum(rate(synapse_util_caches_cache{job=\"$job\",index=~\"$index\", name=\"room_push_rule_cache\",instance=\"$instance\"}[$bucket_size]))", "format": "time_series", "interval": "", "intervalFactor": 2, "legendFormat": "Hit Rate", - "metric": "synapse_push_bulk_push_rule_evaluator_push_rules_invalidation_counter", + "metric": "synapse_push_bulk_push_rule_evaluator_push_rules_invalidation_counter_total", "refId": "A", "step": 2 }, @@ -5613,7 +5613,7 @@ "uid": "${DS_PROMETHEUS}" }, "exemplar": true, - "expr": "sum(rate(synapse_util_caches_cache:total{job=\"$job\",index=~\"$index\", name=\"room_push_rule_cache\",instance=\"$instance\"}[$bucket_size]))", + "expr": "sum(rate(synapse_util_caches_cache{job=\"$job\",index=~\"$index\", name=\"room_push_rule_cache\",instance=\"$instance\"}[$bucket_size]))", "format": "time_series", "interval": "", "intervalFactor": 2, @@ -5719,12 +5719,12 @@ "uid": "${DS_PROMETHEUS}" }, "exemplar": true, - "expr": "sum(rate(synapse_util_caches_cache:hits{job=\"$job\",index=~\"$index\",name=\"_get_rules_for_room\",instance=\"$instance\"}[$bucket_size]))/sum(rate(synapse_util_caches_cache:total{job=\"$job\",index=~\"$index\", name=\"_get_rules_for_room\",instance=\"$instance\"}[$bucket_size]))", + "expr": "sum(rate(synapse_util_caches_cache_hits{job=\"$job\",index=~\"$index\",name=\"_get_rules_for_room\",instance=\"$instance\"}[$bucket_size]))/sum(rate(synapse_util_caches_cache{job=\"$job\",index=~\"$index\", name=\"_get_rules_for_room\",instance=\"$instance\"}[$bucket_size]))", "format": "time_series", "interval": "", "intervalFactor": 2, "legendFormat": "Hit Rate", - "metric": "synapse_push_bulk_push_rule_evaluator_push_rules_invalidation_counter", + "metric": "synapse_push_bulk_push_rule_evaluator_push_rules_invalidation_counter_total", "refId": "A", "step": 2 }, @@ -5734,7 +5734,7 @@ "uid": "${DS_PROMETHEUS}" }, "exemplar": true, - "expr": "sum(rate(synapse_util_caches_cache:total{job=\"$job\",index=~\"$index\", name=\"_get_rules_for_room\",instance=\"$instance\"}[$bucket_size]))", + "expr": "sum(rate(synapse_util_caches_cache{job=\"$job\",index=~\"$index\", name=\"_get_rules_for_room\",instance=\"$instance\"}[$bucket_size]))", "format": "time_series", "interval": "", "intervalFactor": 2, @@ -6087,7 +6087,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "topk(10, rate(synapse_storage_transaction_time_count{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size]))", + "expr": "topk(10, rate(synapse_storage_transaction_time_count_total{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size]))", "format": "time_series", "interval": "", "intervalFactor": 2, @@ -6187,7 +6187,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "rate(synapse_storage_transaction_time_sum{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size])", + "expr": "rate(synapse_storage_transaction_time_sum_total{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size])", "format": "time_series", "instant": false, "interval": "", @@ -6287,7 +6287,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "rate(synapse_storage_transaction_time_sum{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size])/rate(synapse_storage_transaction_time_count{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size])", + "expr": "rate(synapse_storage_transaction_time_sum_total{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size])/rate(synapse_storage_transaction_time_count_total{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size])", "format": "time_series", "instant": false, "interval": "", @@ -6538,7 +6538,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "rate(synapse_util_metrics_block_ru_utime_seconds{instance=\"$instance\",job=~\"$job\",index=~\"$index\",block_name!=\"wrapped_request_handler\"}[$bucket_size]) + rate(synapse_util_metrics_block_ru_stime_seconds[$bucket_size])", + "expr": "rate(synapse_util_metrics_block_ru_utime_seconds_total{instance=\"$instance\",job=~\"$job\",index=~\"$index\",block_name!=\"wrapped_request_handler\"}[$bucket_size]) + rate(synapse_util_metrics_block_ru_stime_seconds_total[$bucket_size])", "format": "time_series", "interval": "", "intervalFactor": 2, @@ -6636,7 +6636,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "(rate(synapse_util_metrics_block_ru_utime_seconds{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size]) + rate(synapse_util_metrics_block_ru_stime_seconds[$bucket_size])) / rate(synapse_util_metrics_block_count[$bucket_size])", + "expr": "(rate(synapse_util_metrics_block_ru_utime_seconds_total{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size]) + rate(synapse_util_metrics_block_ru_stime_seconds_total[$bucket_size])) / rate(synapse_util_metrics_block_count_total[$bucket_size])", "format": "time_series", "interval": "", "intervalFactor": 2, @@ -6737,7 +6737,7 @@ "uid": "${DS_PROMETHEUS}" }, "exemplar": true, - "expr": "rate(synapse_util_metrics_block_db_txn_duration_seconds{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size])", + "expr": "rate(synapse_util_metrics_block_db_txn_duration_seconds_total{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size])", "format": "time_series", "interval": "", "intervalFactor": 2, @@ -6839,7 +6839,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "rate(synapse_util_metrics_block_db_txn_duration_seconds{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size]) / rate(synapse_util_metrics_block_db_txn_count{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size])", + "expr": "rate(synapse_util_metrics_block_db_txn_duration_seconds_total{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size]) / rate(synapse_util_metrics_block_db_txn_count_total{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size])", "format": "time_series", "interval": "", "intervalFactor": 2, @@ -6936,7 +6936,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "rate(synapse_util_metrics_block_db_txn_duration_seconds{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size]) / rate(synapse_util_metrics_block_db_txn_count{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size])", + "expr": "rate(synapse_util_metrics_block_db_txn_duration_seconds_total{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size]) / rate(synapse_util_metrics_block_db_txn_count_total{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size])", "format": "time_series", "interval": "", "intervalFactor": 2, @@ -7033,7 +7033,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "rate(synapse_util_metrics_block_time_seconds{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size]) / rate(synapse_util_metrics_block_count{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size])", + "expr": "rate(synapse_util_metrics_block_time_seconds_total{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size]) / rate(synapse_util_metrics_block_count_total{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size])", "format": "time_series", "interval": "", "intervalFactor": 2, @@ -7122,7 +7122,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "rate(synapse_util_metrics_block_count{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size])", + "expr": "rate(synapse_util_metrics_block_count_total{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size])", "interval": "", "legendFormat": "{{job}}-{{index}} {{block_name}}", "refId": "A" @@ -7246,7 +7246,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "rate(synapse_util_caches_cache:hits{job=~\"$job\",index=~\"$index\",instance=\"$instance\"}[$bucket_size])/rate(synapse_util_caches_cache:total{job=~\"$job\",index=~\"$index\",instance=\"$instance\"}[$bucket_size])", + "expr": "rate(synapse_util_caches_cache_hits{job=~\"$job\",index=~\"$index\",instance=\"$instance\"}[$bucket_size])/rate(synapse_util_caches_cache{job=~\"$job\",index=~\"$index\",instance=\"$instance\"}[$bucket_size])", "format": "time_series", "intervalFactor": 2, "legendFormat": "{{name}} {{job}}-{{index}}", @@ -7347,7 +7347,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "synapse_util_caches_cache:size{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}", + "expr": "synapse_util_caches_cache_size{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}", "format": "time_series", "hide": false, "interval": "", @@ -7447,7 +7447,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "rate(synapse_util_caches_cache:total{job=~\"$job\",index=~\"$index\",instance=\"$instance\"}[$bucket_size])", + "expr": "rate(synapse_util_caches_cache{job=~\"$job\",index=~\"$index\",instance=\"$instance\"}[$bucket_size])", "format": "time_series", "interval": "", "intervalFactor": 2, @@ -7547,7 +7547,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "topk(10, rate(synapse_util_caches_cache:total{job=~\"$job\",index=~\"$index\",instance=\"$instance\"}[$bucket_size]) - rate(synapse_util_caches_cache:hits{job=~\"$job\",index=~\"$index\",instance=\"$instance\"}[$bucket_size]))", + "expr": "topk(10, rate(synapse_util_caches_cache{job=~\"$job\",index=~\"$index\",instance=\"$instance\"}[$bucket_size]) - rate(synapse_util_caches_cache_hits{job=~\"$job\",index=~\"$index\",instance=\"$instance\"}[$bucket_size]))", "format": "time_series", "interval": "", "intervalFactor": 2, @@ -7643,7 +7643,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "rate(synapse_util_caches_cache:evicted_size{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size])", + "expr": "rate(synapse_util_caches_cache_evicted_size{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size])", "format": "time_series", "interval": "", "intervalFactor": 1, @@ -7763,7 +7763,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "synapse_util_caches_response_cache:size{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}", + "expr": "synapse_util_caches_response_cache_size{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}", "interval": "", "legendFormat": "{{name}} {{job}}-{{index}}", "refId": "A" @@ -7853,7 +7853,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "rate(synapse_util_caches_response_cache:hits{instance=\"$instance\", job=~\"$job\", index=~\"$index\"}[$bucket_size])/rate(synapse_util_caches_response_cache:total{instance=\"$instance\", job=~\"$job\", index=~\"$index\"}[$bucket_size])", + "expr": "rate(synapse_util_caches_response_cache_hits{instance=\"$instance\", job=~\"$job\", index=~\"$index\"}[$bucket_size])/rate(synapse_util_caches_response_cache{instance=\"$instance\", job=~\"$job\", index=~\"$index\"}[$bucket_size])", "interval": "", "legendFormat": "{{name}} {{job}}-{{index}}", "refId": "A" @@ -9556,7 +9556,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "synapse_forward_extremities_bucket{instance=\"$instance\"} and on (index, instance, job) (synapse_storage_events_persisted_events > 0)", + "expr": "synapse_forward_extremities_bucket{instance=\"$instance\"} and on (index, instance, job) (synapse_storage_events_persisted_events_total > 0)", "format": "heatmap", "intervalFactor": 1, "legendFormat": "{{le}}", @@ -9716,7 +9716,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "rate(synapse_storage_events_forward_extremities_persisted_bucket{instance=\"$instance\"}[$bucket_size]) and on (index, instance, job) (synapse_storage_events_persisted_events > 0)", + "expr": "rate(synapse_storage_events_forward_extremities_persisted_bucket{instance=\"$instance\"}[$bucket_size]) and on (index, instance, job) (synapse_storage_events_persisted_events_total > 0)", "format": "heatmap", "intervalFactor": 1, "legendFormat": "{{le}}", @@ -9793,7 +9793,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "histogram_quantile(0.5, rate(synapse_storage_events_forward_extremities_persisted_bucket{instance=\"$instance\"}[$bucket_size]) and on (index, instance, job) (synapse_storage_events_persisted_events > 0))", + "expr": "histogram_quantile(0.5, rate(synapse_storage_events_forward_extremities_persisted_bucket{instance=\"$instance\"}[$bucket_size]) and on (index, instance, job) (synapse_storage_events_persisted_events_total > 0))", "format": "time_series", "intervalFactor": 1, "legendFormat": "50%", @@ -9803,7 +9803,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "histogram_quantile(0.75, rate(synapse_storage_events_forward_extremities_persisted_bucket{instance=\"$instance\"}[$bucket_size]) and on (index, instance, job) (synapse_storage_events_persisted_events > 0))", + "expr": "histogram_quantile(0.75, rate(synapse_storage_events_forward_extremities_persisted_bucket{instance=\"$instance\"}[$bucket_size]) and on (index, instance, job) (synapse_storage_events_persisted_events_total > 0))", "format": "time_series", "intervalFactor": 1, "legendFormat": "75%", @@ -9813,7 +9813,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "histogram_quantile(0.90, rate(synapse_storage_events_forward_extremities_persisted_bucket{instance=\"$instance\"}[$bucket_size]) and on (index, instance, job) (synapse_storage_events_persisted_events > 0))", + "expr": "histogram_quantile(0.90, rate(synapse_storage_events_forward_extremities_persisted_bucket{instance=\"$instance\"}[$bucket_size]) and on (index, instance, job) (synapse_storage_events_persisted_events_total > 0))", "format": "time_series", "intervalFactor": 1, "legendFormat": "90%", @@ -9823,7 +9823,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "histogram_quantile(0.99, rate(synapse_storage_events_forward_extremities_persisted_bucket{instance=\"$instance\"}[$bucket_size]) and on (index, instance, job) (synapse_storage_events_persisted_events > 0))", + "expr": "histogram_quantile(0.99, rate(synapse_storage_events_forward_extremities_persisted_bucket{instance=\"$instance\"}[$bucket_size]) and on (index, instance, job) (synapse_storage_events_persisted_events_total > 0))", "format": "time_series", "intervalFactor": 1, "legendFormat": "99%", @@ -9905,7 +9905,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "rate(synapse_storage_events_stale_forward_extremities_persisted_bucket{instance=\"$instance\"}[$bucket_size]) and on (index, instance, job) (synapse_storage_events_persisted_events > 0)", + "expr": "rate(synapse_storage_events_stale_forward_extremities_persisted_bucket{instance=\"$instance\"}[$bucket_size]) and on (index, instance, job) (synapse_storage_events_persisted_events_total > 0)", "format": "heatmap", "intervalFactor": 1, "legendFormat": "{{le}}", @@ -9982,7 +9982,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "histogram_quantile(0.5, rate(synapse_storage_events_stale_forward_extremities_persisted_bucket{instance=\"$instance\"}[$bucket_size]) and on (index, instance, job) (synapse_storage_events_persisted_events > 0))", + "expr": "histogram_quantile(0.5, rate(synapse_storage_events_stale_forward_extremities_persisted_bucket{instance=\"$instance\"}[$bucket_size]) and on (index, instance, job) (synapse_storage_events_persisted_events_total > 0))", "format": "time_series", "intervalFactor": 1, "legendFormat": "50%", @@ -9992,7 +9992,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "histogram_quantile(0.75, rate(synapse_storage_events_stale_forward_extremities_persisted_bucket{instance=\"$instance\"}[$bucket_size]) and on (index, instance, job) (synapse_storage_events_persisted_events > 0))", + "expr": "histogram_quantile(0.75, rate(synapse_storage_events_stale_forward_extremities_persisted_bucket{instance=\"$instance\"}[$bucket_size]) and on (index, instance, job) (synapse_storage_events_persisted_events_total > 0))", "format": "time_series", "intervalFactor": 1, "legendFormat": "75%", @@ -10002,7 +10002,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "histogram_quantile(0.90, rate(synapse_storage_events_stale_forward_extremities_persisted_bucket{instance=\"$instance\"}[$bucket_size]) and on (index, instance, job) (synapse_storage_events_persisted_events > 0))", + "expr": "histogram_quantile(0.90, rate(synapse_storage_events_stale_forward_extremities_persisted_bucket{instance=\"$instance\"}[$bucket_size]) and on (index, instance, job) (synapse_storage_events_persisted_events_total > 0))", "format": "time_series", "intervalFactor": 1, "legendFormat": "90%", @@ -10012,7 +10012,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "histogram_quantile(0.99, rate(synapse_storage_events_stale_forward_extremities_persisted_bucket{instance=\"$instance\"}[$bucket_size]) and on (index, instance, job) (synapse_storage_events_persisted_events > 0))", + "expr": "histogram_quantile(0.99, rate(synapse_storage_events_stale_forward_extremities_persisted_bucket{instance=\"$instance\"}[$bucket_size]) and on (index, instance, job) (synapse_storage_events_persisted_events_total > 0))", "format": "time_series", "intervalFactor": 1, "legendFormat": "99%", @@ -10297,7 +10297,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "sum(rate(synapse_storage_events_state_resolutions_during_persistence{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size]))", + "expr": "sum(rate(synapse_storage_events_state_resolutions_during_persistence_total{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size]))", "interval": "", "legendFormat": "State res ", "refId": "A" @@ -10306,7 +10306,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "sum(rate(synapse_storage_events_potential_times_prune_extremities{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size]))", + "expr": "sum(rate(synapse_storage_events_potential_times_prune_extremities_total{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size]))", "interval": "", "legendFormat": "Potential to prune", "refId": "B" @@ -10315,7 +10315,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "sum(rate(synapse_storage_events_times_pruned_extremities{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size]))", + "expr": "sum(rate(synapse_storage_events_times_pruned_extremities_total{instance=\"$instance\",job=~\"$job\",index=~\"$index\"}[$bucket_size]))", "interval": "", "legendFormat": "Pruned", "refId": "C" @@ -11069,7 +11069,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "rate(synapse_handler_presence_notified_presence{job=\"$job\",index=~\"$index\",instance=\"$instance\"}[$bucket_size])", + "expr": "rate(synapse_handler_presence_notified_presence_total{job=\"$job\",index=~\"$index\",instance=\"$instance\"}[$bucket_size])", "interval": "", "legendFormat": "Notified", "refId": "A" @@ -11078,7 +11078,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "rate(synapse_handler_presence_federation_presence_out{job=\"$job\",index=~\"$index\",instance=\"$instance\"}[$bucket_size])", + "expr": "rate(synapse_handler_presence_federation_presence_out_total{job=\"$job\",index=~\"$index\",instance=\"$instance\"}[$bucket_size])", "interval": "", "legendFormat": "Remote ping", "refId": "B" @@ -11087,7 +11087,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "rate(synapse_handler_presence_presence_updates{job=\"$job\",index=~\"$index\",instance=\"$instance\"}[$bucket_size])", + "expr": "rate(synapse_handler_presence_presence_updates_total{job=\"$job\",index=~\"$index\",instance=\"$instance\"}[$bucket_size])", "interval": "", "legendFormat": "Total updates", "refId": "C" @@ -11096,7 +11096,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "rate(synapse_handler_presence_federation_presence{job=\"$job\",index=~\"$index\",instance=\"$instance\"}[$bucket_size])", + "expr": "rate(synapse_handler_presence_federation_presence_total{job=\"$job\",index=~\"$index\",instance=\"$instance\"}[$bucket_size])", "interval": "", "legendFormat": "Remote updates", "refId": "D" @@ -11105,7 +11105,7 @@ "datasource": { "uid": "$datasource" }, - "expr": "rate(synapse_handler_presence_bump_active_time{job=\"$job\",index=~\"$index\",instance=\"$instance\"}[$bucket_size])", + "expr": "rate(synapse_handler_presence_bump_active_time_total{job=\"$job\",index=~\"$index\",instance=\"$instance\"}[$bucket_size])", "interval": "", "legendFormat": "Bump active time", "refId": "E" @@ -11789,7 +11789,7 @@ "name": "instance", "options": [], "query": { - "query": "label_values(synapse_util_metrics_block_ru_utime_seconds, instance)", + "query": "label_values(synapse_util_metrics_block_ru_utime_seconds_total, instance)", "refId": "Prometheus-instance-Variable-Query" }, "refresh": 2, @@ -11818,7 +11818,7 @@ "name": "job", "options": [], "query": { - "query": "label_values(synapse_util_metrics_block_ru_utime_seconds, job)", + "query": "label_values(synapse_util_metrics_block_ru_utime_seconds_total, job)", "refId": "Prometheus-job-Variable-Query" }, "refresh": 2, @@ -11848,7 +11848,7 @@ "name": "index", "options": [], "query": { - "query": "label_values(synapse_util_metrics_block_ru_utime_seconds, index)", + "query": "label_values(synapse_util_metrics_block_ru_utime_seconds_total, index)", "refId": "Prometheus-index-Variable-Query" }, "refresh": 2, @@ -11896,6 +11896,6 @@ "timezone": "", "title": "Synapse", "uid": "000000012", - "version": 132, + "version": 133, "weekStart": "" -} \ No newline at end of file +} diff --git a/synapse/metrics/_legacy_exposition.py b/synapse/metrics/_legacy_exposition.py index ff640a49af..6f00ff2a47 100644 --- a/synapse/metrics/_legacy_exposition.py +++ b/synapse/metrics/_legacy_exposition.py @@ -88,11 +88,11 @@ LEGACY_METRIC_NAMES = { "synapse_util_caches_cache_hits": "synapse_util_caches_cache:hits", "synapse_util_caches_cache_size": "synapse_util_caches_cache:size", "synapse_util_caches_cache_evicted_size": "synapse_util_caches_cache:evicted_size", - "synapse_util_caches_cache_total": "synapse_util_caches_cache:total", + "synapse_util_caches_cache": "synapse_util_caches_cache:total", "synapse_util_caches_response_cache_size": "synapse_util_caches_response_cache:size", "synapse_util_caches_response_cache_hits": "synapse_util_caches_response_cache:hits", "synapse_util_caches_response_cache_evicted_size": "synapse_util_caches_response_cache:evicted_size", - "synapse_util_caches_response_cache_total": "synapse_util_caches_response_cache:total", + "synapse_util_caches_response_cache": "synapse_util_caches_response_cache:total", } diff --git a/synapse/util/caches/__init__.py b/synapse/util/caches/__init__.py index bdf9b0dc8c..d4a2b77c29 100644 --- a/synapse/util/caches/__init__.py +++ b/synapse/util/caches/__init__.py @@ -37,7 +37,7 @@ collectors_by_name: Dict[str, "CacheMetric"] = {} cache_size = Gauge("synapse_util_caches_cache_size", "", ["name"]) cache_hits = Gauge("synapse_util_caches_cache_hits", "", ["name"]) cache_evicted = Gauge("synapse_util_caches_cache_evicted_size", "", ["name", "reason"]) -cache_total = Gauge("synapse_util_caches_cache_total", "", ["name"]) +cache_total = Gauge("synapse_util_caches_cache", "", ["name"]) cache_max_size = Gauge("synapse_util_caches_cache_max_size", "", ["name"]) cache_memory_usage = Gauge( "synapse_util_caches_cache_size_bytes", @@ -50,7 +50,7 @@ response_cache_hits = Gauge("synapse_util_caches_response_cache_hits", "", ["nam response_cache_evicted = Gauge( "synapse_util_caches_response_cache_evicted_size", "", ["name", "reason"] ) -response_cache_total = Gauge("synapse_util_caches_response_cache_total", "", ["name"]) +response_cache_total = Gauge("synapse_util_caches_response_cache", "", ["name"]) class EvictionReason(Enum): -- cgit 1.5.1 From cf11919ddd4f48b2f59062542ba62969042f80aa Mon Sep 17 00:00:00 2001 From: reivilibre Date: Thu, 8 Sep 2022 14:30:48 +0000 Subject: Fix cache metrics not being updated when not using the legacy exposition module. (#13717) --- changelog.d/13717.misc | 1 + synapse/metrics/_legacy_exposition.py | 7 ---- synapse/util/caches/__init__.py | 60 ++++++++++++++++++++++++++++------- synapse/util/metrics.py | 34 ++++++++++++++++++-- 4 files changed, 81 insertions(+), 21 deletions(-) create mode 100644 changelog.d/13717.misc (limited to 'synapse/util') diff --git a/changelog.d/13717.misc b/changelog.d/13717.misc new file mode 100644 index 0000000000..07ace50b12 --- /dev/null +++ b/changelog.d/13717.misc @@ -0,0 +1 @@ +Add experimental configuration option to allow disabling legacy Prometheus metric names. \ No newline at end of file diff --git a/synapse/metrics/_legacy_exposition.py b/synapse/metrics/_legacy_exposition.py index 133f1603dd..563d8cc2c6 100644 --- a/synapse/metrics/_legacy_exposition.py +++ b/synapse/metrics/_legacy_exposition.py @@ -34,8 +34,6 @@ from prometheus_client.core import Sample from twisted.web.resource import Resource from twisted.web.server import Request -from synapse.util import caches - CONTENT_TYPE_LATEST = "text/plain; version=0.0.4; charset=utf-8" @@ -107,11 +105,6 @@ def generate_latest(registry: CollectorRegistry, emit_help: bool = False) -> byt by prometheus-client. """ - # Trigger the cache metrics to be rescraped, which updates the common - # metrics but do not produce metrics themselves - for collector in caches.collectors_by_name.values(): - collector.collect() - output = [] for metric in registry.collect(): diff --git a/synapse/util/caches/__init__.py b/synapse/util/caches/__init__.py index d4a2b77c29..35c0be08b0 100644 --- a/synapse/util/caches/__init__.py +++ b/synapse/util/caches/__init__.py @@ -20,9 +20,11 @@ from sys import intern from typing import Any, Callable, Dict, List, Optional, Sized, TypeVar import attr +from prometheus_client import REGISTRY from prometheus_client.core import Gauge from synapse.config.cache import add_resizable_cache +from synapse.util.metrics import DynamicCollectorRegistry logger = logging.getLogger(__name__) @@ -30,27 +32,62 @@ logger = logging.getLogger(__name__) # Whether to track estimated memory usage of the LruCaches. TRACK_MEMORY_USAGE = False +# We track cache metrics in a special registry that lets us update the metrics +# just before they are returned from the scrape endpoint. +CACHE_METRIC_REGISTRY = DynamicCollectorRegistry() caches_by_name: Dict[str, Sized] = {} -collectors_by_name: Dict[str, "CacheMetric"] = {} -cache_size = Gauge("synapse_util_caches_cache_size", "", ["name"]) -cache_hits = Gauge("synapse_util_caches_cache_hits", "", ["name"]) -cache_evicted = Gauge("synapse_util_caches_cache_evicted_size", "", ["name", "reason"]) -cache_total = Gauge("synapse_util_caches_cache", "", ["name"]) -cache_max_size = Gauge("synapse_util_caches_cache_max_size", "", ["name"]) +cache_size = Gauge( + "synapse_util_caches_cache_size", "", ["name"], registry=CACHE_METRIC_REGISTRY +) +cache_hits = Gauge( + "synapse_util_caches_cache_hits", "", ["name"], registry=CACHE_METRIC_REGISTRY +) +cache_evicted = Gauge( + "synapse_util_caches_cache_evicted_size", + "", + ["name", "reason"], + registry=CACHE_METRIC_REGISTRY, +) +cache_total = Gauge( + "synapse_util_caches_cache", "", ["name"], registry=CACHE_METRIC_REGISTRY +) +cache_max_size = Gauge( + "synapse_util_caches_cache_max_size", "", ["name"], registry=CACHE_METRIC_REGISTRY +) cache_memory_usage = Gauge( "synapse_util_caches_cache_size_bytes", "Estimated memory usage of the caches", ["name"], + registry=CACHE_METRIC_REGISTRY, ) -response_cache_size = Gauge("synapse_util_caches_response_cache_size", "", ["name"]) -response_cache_hits = Gauge("synapse_util_caches_response_cache_hits", "", ["name"]) +response_cache_size = Gauge( + "synapse_util_caches_response_cache_size", + "", + ["name"], + registry=CACHE_METRIC_REGISTRY, +) +response_cache_hits = Gauge( + "synapse_util_caches_response_cache_hits", + "", + ["name"], + registry=CACHE_METRIC_REGISTRY, +) response_cache_evicted = Gauge( - "synapse_util_caches_response_cache_evicted_size", "", ["name", "reason"] + "synapse_util_caches_response_cache_evicted_size", + "", + ["name", "reason"], + registry=CACHE_METRIC_REGISTRY, ) -response_cache_total = Gauge("synapse_util_caches_response_cache", "", ["name"]) +response_cache_total = Gauge( + "synapse_util_caches_response_cache", "", ["name"], registry=CACHE_METRIC_REGISTRY +) + + +# Register our custom cache metrics registry with the global registry +REGISTRY.register(CACHE_METRIC_REGISTRY) class EvictionReason(Enum): @@ -168,9 +205,8 @@ def register_cache( add_resizable_cache(cache_name, resize_callback) metric = CacheMetric(cache, cache_type, cache_name, collect_callback) - metric_name = "cache_%s_%s" % (cache_type, cache_name) caches_by_name[cache_name] = cache - collectors_by_name[metric_name] = metric + CACHE_METRIC_REGISTRY.register_hook(metric.collect) return metric diff --git a/synapse/util/metrics.py b/synapse/util/metrics.py index bc3b4938ea..9687120ebf 100644 --- a/synapse/util/metrics.py +++ b/synapse/util/metrics.py @@ -15,9 +15,9 @@ import logging from functools import wraps from types import TracebackType -from typing import Awaitable, Callable, Optional, Type, TypeVar +from typing import Awaitable, Callable, Generator, List, Optional, Type, TypeVar -from prometheus_client import Counter +from prometheus_client import CollectorRegistry, Counter, Metric from typing_extensions import Concatenate, ParamSpec, Protocol from synapse.logging.context import ( @@ -208,3 +208,33 @@ class Measure: metrics.real_time_sum += duration # TODO: Add other in flight metrics. + + +class DynamicCollectorRegistry(CollectorRegistry): + """ + Custom Prometheus Collector registry that calls a hook first, allowing you + to update metrics on-demand. + + Don't forget to register this registry with the main registry! + """ + + def __init__(self) -> None: + super().__init__() + self._pre_update_hooks: List[Callable[[], None]] = [] + + def collect(self) -> Generator[Metric, None, None]: + """ + Collects metrics, calling pre-update hooks first. + """ + + for pre_update_hook in self._pre_update_hooks: + pre_update_hook() + + yield from super().collect() + + def register_hook(self, hook: Callable[[], None]) -> None: + """ + Registers a hook that is called before metric collection. + """ + + self._pre_update_hooks.append(hook) -- cgit 1.5.1 From ebfeac7c5ded851a2639911ec6adf9d0fcdb029a Mon Sep 17 00:00:00 2001 From: Erik Johnston Date: Mon, 12 Sep 2022 11:03:42 +0100 Subject: Check if Rust lib needs rebuilding. (#13759) This protects against the common mistake of failing to remember to rebuild Rust code after making changes. --- changelog.d/13759.misc | 1 + rust/Cargo.toml | 4 ++ rust/build.rs | 45 ++++++++++++++++++++++ rust/src/lib.rs | 10 ++++- stubs/synapse/synapse_rust.pyi | 1 + synapse/__init__.py | 5 +++ synapse/util/rust.py | 84 ++++++++++++++++++++++++++++++++++++++++++ 7 files changed, 149 insertions(+), 1 deletion(-) create mode 100644 changelog.d/13759.misc create mode 100644 rust/build.rs create mode 100644 synapse/util/rust.py (limited to 'synapse/util') diff --git a/changelog.d/13759.misc b/changelog.d/13759.misc new file mode 100644 index 0000000000..f91c512483 --- /dev/null +++ b/changelog.d/13759.misc @@ -0,0 +1 @@ +Add a check for editable installs if the Rust library needs rebuilding. diff --git a/rust/Cargo.toml b/rust/Cargo.toml index 0a9760cafc..deddf3cec2 100644 --- a/rust/Cargo.toml +++ b/rust/Cargo.toml @@ -19,3 +19,7 @@ name = "synapse.synapse_rust" [dependencies] pyo3 = { version = "0.16.5", features = ["extension-module", "macros", "abi3", "abi3-py37"] } + +[build-dependencies] +blake2 = "0.10.4" +hex = "0.4.3" diff --git a/rust/build.rs b/rust/build.rs new file mode 100644 index 0000000000..2117975e56 --- /dev/null +++ b/rust/build.rs @@ -0,0 +1,45 @@ +//! This build script calculates the hash of all files in the `src/` +//! directory and adds it as an environment variable during build time. +//! +//! This is used so that the python code can detect when the built native module +//! does not match the source in-tree, helping to detect the case where the +//! source has been updated but the library hasn't been rebuilt. + +use std::path::PathBuf; + +use blake2::{Blake2b512, Digest}; + +fn main() -> Result<(), std::io::Error> { + let mut dirs = vec![PathBuf::from("src")]; + + let mut paths = Vec::new(); + while let Some(path) = dirs.pop() { + let mut entries = std::fs::read_dir(path)? + .map(|res| res.map(|e| e.path())) + .collect::, std::io::Error>>()?; + + entries.sort(); + + for entry in entries { + if entry.is_dir() { + dirs.push(entry) + } else { + paths.push(entry.to_str().expect("valid rust paths").to_string()); + } + } + } + + paths.sort(); + + let mut hasher = Blake2b512::new(); + + for path in paths { + let bytes = std::fs::read(path)?; + hasher.update(bytes); + } + + let hex_digest = hex::encode(hasher.finalize()); + println!("cargo:rustc-env=SYNAPSE_RUST_DIGEST={hex_digest}"); + + Ok(()) +} diff --git a/rust/src/lib.rs b/rust/src/lib.rs index 142fc2ed93..ba42465fb8 100644 --- a/rust/src/lib.rs +++ b/rust/src/lib.rs @@ -1,5 +1,13 @@ use pyo3::prelude::*; +/// Returns the hash of all the rust source files at the time it was compiled. +/// +/// Used by python to detect if the rust library is outdated. +#[pyfunction] +fn get_rust_file_digest() -> &'static str { + env!("SYNAPSE_RUST_DIGEST") +} + /// Formats the sum of two numbers as string. #[pyfunction] #[pyo3(text_signature = "(a, b, /)")] @@ -11,6 +19,6 @@ fn sum_as_string(a: usize, b: usize) -> PyResult { #[pymodule] fn synapse_rust(_py: Python<'_>, m: &PyModule) -> PyResult<()> { m.add_function(wrap_pyfunction!(sum_as_string, m)?)?; - + m.add_function(wrap_pyfunction!(get_rust_file_digest, m)?)?; Ok(()) } diff --git a/stubs/synapse/synapse_rust.pyi b/stubs/synapse/synapse_rust.pyi index 5b51ba05d7..8658d3138f 100644 --- a/stubs/synapse/synapse_rust.pyi +++ b/stubs/synapse/synapse_rust.pyi @@ -1 +1,2 @@ def sum_as_string(a: int, b: int) -> str: ... +def get_rust_file_digest() -> str: ... diff --git a/synapse/__init__.py b/synapse/__init__.py index b1369aca8f..1bed6393bd 100644 --- a/synapse/__init__.py +++ b/synapse/__init__.py @@ -20,6 +20,8 @@ import json import os import sys +from synapse.util.rust import check_rust_lib_up_to_date + # Check that we're not running on an unsupported Python version. if sys.version_info < (3, 7): print("Synapse requires Python 3.7 or above.") @@ -78,3 +80,6 @@ if bool(os.environ.get("SYNAPSE_TEST_PATCH_LOG_CONTEXTS", False)): from synapse.util.patch_inline_callbacks import do_patch do_patch() + + +check_rust_lib_up_to_date() diff --git a/synapse/util/rust.py b/synapse/util/rust.py new file mode 100644 index 0000000000..30ecb9ffd9 --- /dev/null +++ b/synapse/util/rust.py @@ -0,0 +1,84 @@ +# Copyright 2022 The Matrix.org Foundation C.I.C. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +import os +import sys +from hashlib import blake2b + +import synapse +from synapse.synapse_rust import get_rust_file_digest + + +def check_rust_lib_up_to_date() -> None: + """For editable installs check if the rust library is outdated and needs to + be rebuilt. + """ + + if not _dist_is_editable(): + return + + synapse_dir = os.path.dirname(synapse.__file__) + synapse_root = os.path.abspath(os.path.join(synapse_dir, "..")) + + # Double check we've not gone into site-packages... + if os.path.basename(synapse_root) == "site-packages": + return + + # ... and it looks like the root of a python project. + if not os.path.exists("pyproject.toml"): + return + + # Get the hash of all Rust source files + hash = _hash_rust_files_in_directory(os.path.join(synapse_root, "rust", "src")) + + if hash != get_rust_file_digest(): + raise Exception("Rust module outdated. Please rebuild using `poetry install`") + + +def _hash_rust_files_in_directory(directory: str) -> str: + """Get the hash of all files in a directory (recursively)""" + + directory = os.path.abspath(directory) + + paths = [] + + dirs = [directory] + while dirs: + dir = dirs.pop() + with os.scandir(dir) as d: + for entry in d: + if entry.is_dir(): + dirs.append(entry.path) + else: + paths.append(entry.path) + + # We sort to make sure that we get a consistent and well-defined ordering. + paths.sort() + + hasher = blake2b() + + for path in paths: + with open(os.path.join(directory, path), "rb") as f: + hasher.update(f.read()) + + return hasher.hexdigest() + + +def _dist_is_editable() -> bool: + """Is distribution an editable install?""" + for path_item in sys.path: + egg_link = os.path.join(path_item, "matrix-synapse.egg-link") + if os.path.isfile(egg_link): + return True + return False -- cgit 1.5.1 From cf65433de26ecce551c64e56d9ee8435c99defab Mon Sep 17 00:00:00 2001 From: reivilibre Date: Wed, 14 Sep 2022 15:29:05 +0000 Subject: Fix a memory leak when running the unit tests. (#13798) --- changelog.d/13798.misc | 1 + synapse/util/caches/__init__.py | 3 ++- synapse/util/metrics.py | 10 +++++----- 3 files changed, 8 insertions(+), 6 deletions(-) create mode 100644 changelog.d/13798.misc (limited to 'synapse/util') diff --git a/changelog.d/13798.misc b/changelog.d/13798.misc new file mode 100644 index 0000000000..e4ec2d77d6 --- /dev/null +++ b/changelog.d/13798.misc @@ -0,0 +1 @@ +Fix a memory leak when running the unit tests. \ No newline at end of file diff --git a/synapse/util/caches/__init__.py b/synapse/util/caches/__init__.py index 35c0be08b0..f7c3a6794e 100644 --- a/synapse/util/caches/__init__.py +++ b/synapse/util/caches/__init__.py @@ -205,8 +205,9 @@ def register_cache( add_resizable_cache(cache_name, resize_callback) metric = CacheMetric(cache, cache_type, cache_name, collect_callback) + metric_name = "cache_%s_%s" % (cache_type, cache_name) caches_by_name[cache_name] = cache - CACHE_METRIC_REGISTRY.register_hook(metric.collect) + CACHE_METRIC_REGISTRY.register_hook(metric_name, metric.collect) return metric diff --git a/synapse/util/metrics.py b/synapse/util/metrics.py index 9687120ebf..165480bdbe 100644 --- a/synapse/util/metrics.py +++ b/synapse/util/metrics.py @@ -15,7 +15,7 @@ import logging from functools import wraps from types import TracebackType -from typing import Awaitable, Callable, Generator, List, Optional, Type, TypeVar +from typing import Awaitable, Callable, Dict, Generator, Optional, Type, TypeVar from prometheus_client import CollectorRegistry, Counter, Metric from typing_extensions import Concatenate, ParamSpec, Protocol @@ -220,21 +220,21 @@ class DynamicCollectorRegistry(CollectorRegistry): def __init__(self) -> None: super().__init__() - self._pre_update_hooks: List[Callable[[], None]] = [] + self._pre_update_hooks: Dict[str, Callable[[], None]] = {} def collect(self) -> Generator[Metric, None, None]: """ Collects metrics, calling pre-update hooks first. """ - for pre_update_hook in self._pre_update_hooks: + for pre_update_hook in self._pre_update_hooks.values(): pre_update_hook() yield from super().collect() - def register_hook(self, hook: Callable[[], None]) -> None: + def register_hook(self, metric_name: str, hook: Callable[[], None]) -> None: """ Registers a hook that is called before metric collection. """ - self._pre_update_hooks.append(hook) + self._pre_update_hooks[metric_name] = hook -- cgit 1.5.1 From 6bd8763804dc0987c7ecd37bcb5ebff465fffa29 Mon Sep 17 00:00:00 2001 From: Mathieu Velten Date: Wed, 21 Sep 2022 15:32:01 +0200 Subject: Add cache invalidation across workers to module API (#13667) Signed-off-by: Mathieu Velten --- changelog.d/13667.feature | 1 + scripts-dev/mypy_synapse_plugin.py | 4 +- synapse/module_api/__init__.py | 33 ++++++++- synapse/storage/_base.py | 23 +++++-- synapse/storage/databases/main/cache.py | 20 ++++-- synapse/util/caches/descriptors.py | 14 ++-- .../replication/test_module_cache_invalidation.py | 79 ++++++++++++++++++++++ 7 files changed, 153 insertions(+), 21 deletions(-) create mode 100644 changelog.d/13667.feature create mode 100644 tests/replication/test_module_cache_invalidation.py (limited to 'synapse/util') diff --git a/changelog.d/13667.feature b/changelog.d/13667.feature new file mode 100644 index 0000000000..a0b3cfe18c --- /dev/null +++ b/changelog.d/13667.feature @@ -0,0 +1 @@ +Add cache invalidation across workers to module API. diff --git a/scripts-dev/mypy_synapse_plugin.py b/scripts-dev/mypy_synapse_plugin.py index d08517a953..2c377533c0 100644 --- a/scripts-dev/mypy_synapse_plugin.py +++ b/scripts-dev/mypy_synapse_plugin.py @@ -29,7 +29,7 @@ class SynapsePlugin(Plugin): self, fullname: str ) -> Optional[Callable[[MethodSigContext], CallableType]]: if fullname.startswith( - "synapse.util.caches.descriptors._CachedFunction.__call__" + "synapse.util.caches.descriptors.CachedFunction.__call__" ) or fullname.startswith( "synapse.util.caches.descriptors._LruCachedFunction.__call__" ): @@ -38,7 +38,7 @@ class SynapsePlugin(Plugin): def cached_function_method_signature(ctx: MethodSigContext) -> CallableType: - """Fixes the `_CachedFunction.__call__` signature to be correct. + """Fixes the `CachedFunction.__call__` signature to be correct. It already has *almost* the correct signature, except: diff --git a/synapse/module_api/__init__.py b/synapse/module_api/__init__.py index 9287c0fb8d..59755bff6d 100644 --- a/synapse/module_api/__init__.py +++ b/synapse/module_api/__init__.py @@ -125,7 +125,7 @@ from synapse.types import ( ) from synapse.util import Clock from synapse.util.async_helpers import maybe_awaitable -from synapse.util.caches.descriptors import cached +from synapse.util.caches.descriptors import CachedFunction, cached from synapse.util.frozenutils import freeze if TYPE_CHECKING: @@ -836,6 +836,37 @@ class ModuleApi: self._store.db_pool.runInteraction(desc, func, *args, **kwargs) # type: ignore[arg-type] ) + def register_cached_function(self, cached_func: CachedFunction) -> None: + """Register a cached function that should be invalidated across workers. + Invalidation local to a worker can be done directly using `cached_func.invalidate`, + however invalidation that needs to go to other workers needs to call `invalidate_cache` + on the module API instead. + + Args: + cached_function: The cached function that will be registered to receive invalidation + locally and from other workers. + """ + self._store.register_external_cached_function( + f"{cached_func.__module__}.{cached_func.__name__}", cached_func + ) + + async def invalidate_cache( + self, cached_func: CachedFunction, keys: Tuple[Any, ...] + ) -> None: + """Invalidate a cache entry of a cached function across workers. The cached function + needs to be registered on all workers first with `register_cached_function`. + + Args: + cached_function: The cached function that needs an invalidation + keys: keys of the entry to invalidate, usually matching the arguments of the + cached function. + """ + cached_func.invalidate(keys) + await self._store.send_invalidation_to_replication( + f"{cached_func.__module__}.{cached_func.__name__}", + keys, + ) + async def complete_sso_login_async( self, registered_user_id: str, diff --git a/synapse/storage/_base.py b/synapse/storage/_base.py index e30f9c76d4..303a5d5298 100644 --- a/synapse/storage/_base.py +++ b/synapse/storage/_base.py @@ -15,12 +15,13 @@ # limitations under the License. import logging from abc import ABCMeta -from typing import TYPE_CHECKING, Any, Collection, Iterable, Optional, Union +from typing import TYPE_CHECKING, Any, Collection, Dict, Iterable, Optional, Union from synapse.storage.database import make_in_list_sql_clause # noqa: F401; noqa: F401 from synapse.storage.database import DatabasePool, LoggingDatabaseConnection from synapse.types import get_domain_from_id from synapse.util import json_decoder +from synapse.util.caches.descriptors import CachedFunction if TYPE_CHECKING: from synapse.server import HomeServer @@ -47,6 +48,8 @@ class SQLBaseStore(metaclass=ABCMeta): self.database_engine = database.engine self.db_pool = database + self.external_cached_functions: Dict[str, CachedFunction] = {} + def process_replication_rows( self, stream_name: str, @@ -95,7 +98,7 @@ class SQLBaseStore(metaclass=ABCMeta): def _attempt_to_invalidate_cache( self, cache_name: str, key: Optional[Collection[Any]] - ) -> None: + ) -> bool: """Attempts to invalidate the cache of the given name, ignoring if the cache doesn't exist. Mainly used for invalidating caches on workers, where they may not have the cache. @@ -113,9 +116,12 @@ class SQLBaseStore(metaclass=ABCMeta): try: cache = getattr(self, cache_name) except AttributeError: - # We probably haven't pulled in the cache in this worker, - # which is fine. - return + # Check if an externally defined module cache has been registered + cache = self.external_cached_functions.get(cache_name) + if not cache: + # We probably haven't pulled in the cache in this worker, + # which is fine. + return False if key is None: cache.invalidate_all() @@ -125,6 +131,13 @@ class SQLBaseStore(metaclass=ABCMeta): invalidate_method = getattr(cache, "invalidate_local", cache.invalidate) invalidate_method(tuple(key)) + return True + + def register_external_cached_function( + self, cache_name: str, func: CachedFunction + ) -> None: + self.external_cached_functions[cache_name] = func + def db_to_json(db_content: Union[memoryview, bytes, bytearray, str]) -> Any: """ diff --git a/synapse/storage/databases/main/cache.py b/synapse/storage/databases/main/cache.py index 12e9a42382..2c421151c1 100644 --- a/synapse/storage/databases/main/cache.py +++ b/synapse/storage/databases/main/cache.py @@ -33,7 +33,7 @@ from synapse.storage.database import ( ) from synapse.storage.engines import PostgresEngine from synapse.storage.util.id_generators import MultiWriterIdGenerator -from synapse.util.caches.descriptors import _CachedFunction +from synapse.util.caches.descriptors import CachedFunction from synapse.util.iterutils import batch_iter if TYPE_CHECKING: @@ -269,9 +269,7 @@ class CacheInvalidationWorkerStore(SQLBaseStore): return cache_func.invalidate(keys) - await self.db_pool.runInteraction( - "invalidate_cache_and_stream", - self._send_invalidation_to_replication, + await self.send_invalidation_to_replication( cache_func.__name__, keys, ) @@ -279,7 +277,7 @@ class CacheInvalidationWorkerStore(SQLBaseStore): def _invalidate_cache_and_stream( self, txn: LoggingTransaction, - cache_func: _CachedFunction, + cache_func: CachedFunction, keys: Tuple[Any, ...], ) -> None: """Invalidates the cache and adds it to the cache stream so slaves @@ -293,7 +291,7 @@ class CacheInvalidationWorkerStore(SQLBaseStore): self._send_invalidation_to_replication(txn, cache_func.__name__, keys) def _invalidate_all_cache_and_stream( - self, txn: LoggingTransaction, cache_func: _CachedFunction + self, txn: LoggingTransaction, cache_func: CachedFunction ) -> None: """Invalidates the entire cache and adds it to the cache stream so slaves will know to invalidate their caches. @@ -334,6 +332,16 @@ class CacheInvalidationWorkerStore(SQLBaseStore): txn, CURRENT_STATE_CACHE_NAME, [room_id] ) + async def send_invalidation_to_replication( + self, cache_name: str, keys: Optional[Collection[Any]] + ) -> None: + await self.db_pool.runInteraction( + "send_invalidation_to_replication", + self._send_invalidation_to_replication, + cache_name, + keys, + ) + def _send_invalidation_to_replication( self, txn: LoggingTransaction, cache_name: str, keys: Optional[Iterable[Any]] ) -> None: diff --git a/synapse/util/caches/descriptors.py b/synapse/util/caches/descriptors.py index 10aff4d04a..3909f1caea 100644 --- a/synapse/util/caches/descriptors.py +++ b/synapse/util/caches/descriptors.py @@ -53,7 +53,7 @@ CacheKey = Union[Tuple, Any] F = TypeVar("F", bound=Callable[..., Any]) -class _CachedFunction(Generic[F]): +class CachedFunction(Generic[F]): invalidate: Any = None invalidate_all: Any = None prefill: Any = None @@ -242,7 +242,7 @@ class LruCacheDescriptor(_CacheDescriptorBase): return ret2 - wrapped = cast(_CachedFunction, _wrapped) + wrapped = cast(CachedFunction, _wrapped) wrapped.cache = cache obj.__dict__[self.name] = wrapped @@ -363,7 +363,7 @@ class DeferredCacheDescriptor(_CacheDescriptorBase): return make_deferred_yieldable(ret) - wrapped = cast(_CachedFunction, _wrapped) + wrapped = cast(CachedFunction, _wrapped) if self.num_args == 1: assert not self.tree @@ -572,7 +572,7 @@ def cached( iterable: bool = False, prune_unread_entries: bool = True, name: Optional[str] = None, -) -> Callable[[F], _CachedFunction[F]]: +) -> Callable[[F], CachedFunction[F]]: func = lambda orig: DeferredCacheDescriptor( orig, max_entries=max_entries, @@ -585,7 +585,7 @@ def cached( name=name, ) - return cast(Callable[[F], _CachedFunction[F]], func) + return cast(Callable[[F], CachedFunction[F]], func) def cachedList( @@ -594,7 +594,7 @@ def cachedList( list_name: str, num_args: Optional[int] = None, name: Optional[str] = None, -) -> Callable[[F], _CachedFunction[F]]: +) -> Callable[[F], CachedFunction[F]]: """Creates a descriptor that wraps a function in a `DeferredCacheListDescriptor`. Used to do batch lookups for an already created cache. One of the arguments @@ -631,7 +631,7 @@ def cachedList( name=name, ) - return cast(Callable[[F], _CachedFunction[F]], func) + return cast(Callable[[F], CachedFunction[F]], func) def _get_cache_key_builder( diff --git a/tests/replication/test_module_cache_invalidation.py b/tests/replication/test_module_cache_invalidation.py new file mode 100644 index 0000000000..b93cae67d3 --- /dev/null +++ b/tests/replication/test_module_cache_invalidation.py @@ -0,0 +1,79 @@ +# Copyright 2022 The Matrix.org Foundation C.I.C. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. +import logging + +import synapse +from synapse.module_api import cached + +from tests.replication._base import BaseMultiWorkerStreamTestCase + +logger = logging.getLogger(__name__) + +FIRST_VALUE = "one" +SECOND_VALUE = "two" + +KEY = "mykey" + + +class TestCache: + current_value = FIRST_VALUE + + @cached() + async def cached_function(self, user_id: str) -> str: + return self.current_value + + +class ModuleCacheInvalidationTestCase(BaseMultiWorkerStreamTestCase): + servlets = [ + synapse.rest.admin.register_servlets, + ] + + def test_module_cache_full_invalidation(self): + main_cache = TestCache() + self.hs.get_module_api().register_cached_function(main_cache.cached_function) + + worker_hs = self.make_worker_hs("synapse.app.generic_worker") + + worker_cache = TestCache() + worker_hs.get_module_api().register_cached_function( + worker_cache.cached_function + ) + + self.assertEqual(FIRST_VALUE, self.get_success(main_cache.cached_function(KEY))) + self.assertEqual( + FIRST_VALUE, self.get_success(worker_cache.cached_function(KEY)) + ) + + main_cache.current_value = SECOND_VALUE + worker_cache.current_value = SECOND_VALUE + # No invalidation yet, should return the cached value on both the main process and the worker + self.assertEqual(FIRST_VALUE, self.get_success(main_cache.cached_function(KEY))) + self.assertEqual( + FIRST_VALUE, self.get_success(worker_cache.cached_function(KEY)) + ) + + # Full invalidation on the main process, should be replicated on the worker that + # should returned the updated value too + self.get_success( + self.hs.get_module_api().invalidate_cache( + main_cache.cached_function, (KEY,) + ) + ) + + self.assertEqual( + SECOND_VALUE, self.get_success(main_cache.cached_function(KEY)) + ) + self.assertEqual( + SECOND_VALUE, self.get_success(worker_cache.cached_function(KEY)) + ) -- cgit 1.5.1 From 29269d9d3f3419a3d92cdd80dae4a37e2d99a395 Mon Sep 17 00:00:00 2001 From: Eric Eastwood Date: Tue, 27 Sep 2022 15:55:43 -0500 Subject: Fix `have_seen_event` cache not being invalidated (#13863) Fix https://github.com/matrix-org/synapse/issues/13856 Fix https://github.com/matrix-org/synapse/issues/13865 > Discovered while trying to make Synapse fast enough for [this MSC2716 test for importing many batches](https://github.com/matrix-org/complement/pull/214#discussion_r741678240). As an example, disabling the `have_seen_event` cache saves 10 seconds for each `/messages` request in that MSC2716 Complement test because we're not making as many federation requests for `/state` (speeding up `have_seen_event` itself is related to https://github.com/matrix-org/synapse/issues/13625) > > But this will also make `/messages` faster in general so we can include it in the [faster `/messages` milestone](https://github.com/matrix-org/synapse/milestone/11). > > *-- https://github.com/matrix-org/synapse/issues/13856* ### The problem `_invalidate_caches_for_event` doesn't run in monolith mode which means we never even tried to clear the `have_seen_event` and other caches. And even in worker mode, it only runs on the workers, not the master (AFAICT). Additionally there was bug with the key being wrong so `_invalidate_caches_for_event` never invalidates the `have_seen_event` cache even when it does run. Because we were using the `@cachedList` wrong, it was putting items in the cache under keys like `((room_id, event_id),)` with a `set` in a `set` (ex. `(('!TnCIJPKzdQdUlIyXdQ:test', '$Iu0eqEBN7qcyF1S9B3oNB3I91v2o5YOgRNPwi_78s-k'),)`) and we we're trying to invalidate with just `(room_id, event_id)` which did nothing. --- changelog.d/13863.bugfix | 1 + synapse/storage/databases/main/events_worker.py | 40 +++--- synapse/util/caches/descriptors.py | 6 + tests/storage/databases/main/test_events_worker.py | 152 ++++++++++++++------- tests/util/caches/test_descriptors.py | 33 ++++- 5 files changed, 165 insertions(+), 67 deletions(-) create mode 100644 changelog.d/13863.bugfix (limited to 'synapse/util') diff --git a/changelog.d/13863.bugfix b/changelog.d/13863.bugfix new file mode 100644 index 0000000000..74264a4fab --- /dev/null +++ b/changelog.d/13863.bugfix @@ -0,0 +1 @@ +Fix `have_seen_event` cache not being invalidated after we persist an event which causes inefficiency effects like extra `/state` federation calls. diff --git a/synapse/storage/databases/main/events_worker.py b/synapse/storage/databases/main/events_worker.py index 52914febf9..7cdc9fe98f 100644 --- a/synapse/storage/databases/main/events_worker.py +++ b/synapse/storage/databases/main/events_worker.py @@ -1474,32 +1474,38 @@ class EventsWorkerStore(SQLBaseStore): # the batches as big as possible. results: Set[str] = set() - for chunk in batch_iter(event_ids, 500): - r = await self._have_seen_events_dict( - [(room_id, event_id) for event_id in chunk] + for event_ids_chunk in batch_iter(event_ids, 500): + events_seen_dict = await self._have_seen_events_dict( + room_id, event_ids_chunk + ) + results.update( + eid for (eid, have_event) in events_seen_dict.items() if have_event ) - results.update(eid for ((_rid, eid), have_event) in r.items() if have_event) return results - @cachedList(cached_method_name="have_seen_event", list_name="keys") + @cachedList(cached_method_name="have_seen_event", list_name="event_ids") async def _have_seen_events_dict( - self, keys: Collection[Tuple[str, str]] - ) -> Dict[Tuple[str, str], bool]: + self, + room_id: str, + event_ids: Collection[str], + ) -> Dict[str, bool]: """Helper for have_seen_events Returns: - a dict {(room_id, event_id)-> bool} + a dict {event_id -> bool} """ # if the event cache contains the event, obviously we've seen it. cache_results = { - (rid, eid) - for (rid, eid) in keys - if await self._get_event_cache.contains((eid,)) + event_id + for event_id in event_ids + if await self._get_event_cache.contains((event_id,)) } results = dict.fromkeys(cache_results, True) - remaining = [k for k in keys if k not in cache_results] + remaining = [ + event_id for event_id in event_ids if event_id not in cache_results + ] if not remaining: return results @@ -1511,23 +1517,21 @@ class EventsWorkerStore(SQLBaseStore): sql = "SELECT event_id FROM events AS e WHERE " clause, args = make_in_list_sql_clause( - txn.database_engine, "e.event_id", [eid for (_rid, eid) in remaining] + txn.database_engine, "e.event_id", remaining ) txn.execute(sql + clause, args) found_events = {eid for eid, in txn} # ... and then we can update the results for each key - results.update( - {(rid, eid): (eid in found_events) for (rid, eid) in remaining} - ) + results.update({eid: (eid in found_events) for eid in remaining}) await self.db_pool.runInteraction("have_seen_events", have_seen_events_txn) return results @cached(max_entries=100000, tree=True) async def have_seen_event(self, room_id: str, event_id: str) -> bool: - res = await self._have_seen_events_dict(((room_id, event_id),)) - return res[(room_id, event_id)] + res = await self._have_seen_events_dict(room_id, [event_id]) + return res[event_id] def _get_current_state_event_counts_txn( self, txn: LoggingTransaction, room_id: str diff --git a/synapse/util/caches/descriptors.py b/synapse/util/caches/descriptors.py index 3909f1caea..0391966462 100644 --- a/synapse/util/caches/descriptors.py +++ b/synapse/util/caches/descriptors.py @@ -431,6 +431,12 @@ class DeferredCacheListDescriptor(_CacheDescriptorBase): cache: DeferredCache[CacheKey, Any] = cached_method.cache num_args = cached_method.num_args + if num_args != self.num_args: + raise Exception( + "Number of args (%s) does not match underlying cache_method_name=%s (%s)." + % (self.num_args, self.cached_method_name, num_args) + ) + @functools.wraps(self.orig) def wrapped(*args: Any, **kwargs: Any) -> "defer.Deferred[Dict]": # If we're passed a cache_context then we'll want to call its diff --git a/tests/storage/databases/main/test_events_worker.py b/tests/storage/databases/main/test_events_worker.py index 67401272ac..32a798d74b 100644 --- a/tests/storage/databases/main/test_events_worker.py +++ b/tests/storage/databases/main/test_events_worker.py @@ -35,66 +35,45 @@ from synapse.util import Clock from synapse.util.async_helpers import yieldable_gather_results from tests import unittest +from tests.test_utils.event_injection import create_event, inject_event class HaveSeenEventsTestCase(unittest.HomeserverTestCase): + servlets = [ + admin.register_servlets, + room.register_servlets, + login.register_servlets, + ] + def prepare(self, reactor, clock, hs): + self.hs = hs self.store: EventsWorkerStore = hs.get_datastores().main - # insert some test data - for rid in ("room1", "room2"): - self.get_success( - self.store.db_pool.simple_insert( - "rooms", - {"room_id": rid, "room_version": 4}, - ) - ) + self.user = self.register_user("user", "pass") + self.token = self.login(self.user, "pass") + self.room_id = self.helper.create_room_as(self.user, tok=self.token) self.event_ids: List[str] = [] - for idx, rid in enumerate( - ( - "room1", - "room1", - "room1", - "room2", - ) - ): - event_json = {"type": f"test {idx}", "room_id": rid} - event = make_event_from_dict(event_json, room_version=RoomVersions.V4) - event_id = event.event_id - - self.get_success( - self.store.db_pool.simple_insert( - "events", - { - "event_id": event_id, - "room_id": rid, - "topological_ordering": idx, - "stream_ordering": idx, - "type": event.type, - "processed": True, - "outlier": False, - }, + for i in range(3): + event = self.get_success( + inject_event( + hs, + room_version=RoomVersions.V7.identifier, + room_id=self.room_id, + sender=self.user, + type="test_event_type", + content={"body": f"foobarbaz{i}"}, ) ) - self.get_success( - self.store.db_pool.simple_insert( - "event_json", - { - "event_id": event_id, - "room_id": rid, - "json": json.dumps(event_json), - "internal_metadata": "{}", - "format_version": 3, - }, - ) - ) - self.event_ids.append(event_id) + + self.event_ids.append(event.event_id) def test_simple(self): with LoggingContext(name="test") as ctx: res = self.get_success( - self.store.have_seen_events("room1", [self.event_ids[0], "event19"]) + self.store.have_seen_events( + self.room_id, [self.event_ids[0], "eventdoesnotexist"] + ) ) self.assertEqual(res, {self.event_ids[0]}) @@ -104,7 +83,9 @@ class HaveSeenEventsTestCase(unittest.HomeserverTestCase): # a second lookup of the same events should cause no queries with LoggingContext(name="test") as ctx: res = self.get_success( - self.store.have_seen_events("room1", [self.event_ids[0], "event19"]) + self.store.have_seen_events( + self.room_id, [self.event_ids[0], "eventdoesnotexist"] + ) ) self.assertEqual(res, {self.event_ids[0]}) self.assertEqual(ctx.get_resource_usage().db_txn_count, 0) @@ -116,11 +97,86 @@ class HaveSeenEventsTestCase(unittest.HomeserverTestCase): # looking it up should now cause no db hits with LoggingContext(name="test") as ctx: res = self.get_success( - self.store.have_seen_events("room1", [self.event_ids[0]]) + self.store.have_seen_events(self.room_id, [self.event_ids[0]]) ) self.assertEqual(res, {self.event_ids[0]}) self.assertEqual(ctx.get_resource_usage().db_txn_count, 0) + def test_persisting_event_invalidates_cache(self): + """ + Test to make sure that the `have_seen_event` cache + is invalidated after we persist an event and returns + the updated value. + """ + event, event_context = self.get_success( + create_event( + self.hs, + room_id=self.room_id, + sender=self.user, + type="test_event_type", + content={"body": "garply"}, + ) + ) + + with LoggingContext(name="test") as ctx: + # First, check `have_seen_event` for an event we have not seen yet + # to prime the cache with a `false` value. + res = self.get_success( + self.store.have_seen_events(event.room_id, [event.event_id]) + ) + self.assertEqual(res, set()) + + # That should result in a single db query to lookup + self.assertEqual(ctx.get_resource_usage().db_txn_count, 1) + + # Persist the event which should invalidate or prefill the + # `have_seen_event` cache so we don't return stale values. + persistence = self.hs.get_storage_controllers().persistence + self.get_success( + persistence.persist_event( + event, + event_context, + ) + ) + + with LoggingContext(name="test") as ctx: + # Check `have_seen_event` again and we should see the updated fact + # that we have now seen the event after persisting it. + res = self.get_success( + self.store.have_seen_events(event.room_id, [event.event_id]) + ) + self.assertEqual(res, {event.event_id}) + + # That should result in a single db query to lookup + self.assertEqual(ctx.get_resource_usage().db_txn_count, 1) + + def test_invalidate_cache_by_room_id(self): + """ + Test to make sure that all events associated with the given `(room_id,)` + are invalidated in the `have_seen_event` cache. + """ + with LoggingContext(name="test") as ctx: + # Prime the cache with some values + res = self.get_success( + self.store.have_seen_events(self.room_id, self.event_ids) + ) + self.assertEqual(res, set(self.event_ids)) + + # That should result in a single db query to lookup + self.assertEqual(ctx.get_resource_usage().db_txn_count, 1) + + # Clear the cache with any events associated with the `room_id` + self.store.have_seen_event.invalidate((self.room_id,)) + + with LoggingContext(name="test") as ctx: + res = self.get_success( + self.store.have_seen_events(self.room_id, self.event_ids) + ) + self.assertEqual(res, set(self.event_ids)) + + # Since we cleared the cache, it should result in another db query to lookup + self.assertEqual(ctx.get_resource_usage().db_txn_count, 1) + class EventCacheTestCase(unittest.HomeserverTestCase): """Test that the various layers of event cache works.""" diff --git a/tests/util/caches/test_descriptors.py b/tests/util/caches/test_descriptors.py index 48e616ac74..90861fe522 100644 --- a/tests/util/caches/test_descriptors.py +++ b/tests/util/caches/test_descriptors.py @@ -13,7 +13,7 @@ # See the License for the specific language governing permissions and # limitations under the License. import logging -from typing import Set +from typing import Iterable, Set, Tuple from unittest import mock from twisted.internet import defer, reactor @@ -1008,3 +1008,34 @@ class CachedListDescriptorTestCase(unittest.TestCase): obj.inner_context_was_finished, "Tried to restart a finished logcontext" ) self.assertEqual(current_context(), SENTINEL_CONTEXT) + + def test_num_args_mismatch(self): + """ + Make sure someone does not accidentally use @cachedList on a method with + a mismatch in the number args to the underlying single cache method. + """ + + class Cls: + @descriptors.cached(tree=True) + def fn(self, room_id, event_id): + pass + + # This is wrong ❌. `@cachedList` expects to be given the same number + # of arguments as the underlying cached function, just with one of + # the arguments being an iterable + @descriptors.cachedList(cached_method_name="fn", list_name="keys") + def list_fn(self, keys: Iterable[Tuple[str, str]]): + pass + + # Corrected syntax ✅ + # + # @cachedList(cached_method_name="fn", list_name="event_ids") + # async def list_fn( + # self, room_id: str, event_ids: Collection[str], + # ) + + obj = Cls() + + # Make sure this raises an error about the arg mismatch + with self.assertRaises(Exception): + obj.list_fn([("foo", "bar")]) -- cgit 1.5.1 From 6f0c3e669da458e838e7b4b165a13e8a5312d6d0 Mon Sep 17 00:00:00 2001 From: David Robertson Date: Thu, 29 Sep 2022 21:16:08 +0100 Subject: Don't require `setuptools_rust` at runtime (#13952) --- changelog.d/13952.bugfix | 1 + synapse/util/check_dependencies.py | 17 ++++++++++++++++- tests/util/test_check_dependencies.py | 20 ++++++++++++++++++-- 3 files changed, 35 insertions(+), 3 deletions(-) create mode 100644 changelog.d/13952.bugfix (limited to 'synapse/util') diff --git a/changelog.d/13952.bugfix b/changelog.d/13952.bugfix new file mode 100644 index 0000000000..a6af20f051 --- /dev/null +++ b/changelog.d/13952.bugfix @@ -0,0 +1 @@ +Fix a bug introduced in v1.68.0 where Synapse would require `setuptools_rust` at runtime, even though the package is only required at build time. diff --git a/synapse/util/check_dependencies.py b/synapse/util/check_dependencies.py index 66f1da7502..3b1e205700 100644 --- a/synapse/util/check_dependencies.py +++ b/synapse/util/check_dependencies.py @@ -66,6 +66,21 @@ def _is_dev_dependency(req: Requirement) -> bool: ) +def _should_ignore_runtime_requirement(req: Requirement) -> bool: + # This is a build-time dependency. Irritatingly, `poetry build` ignores the + # requirements listed in the [build-system] section of pyproject.toml, so in order + # to support `poetry install --no-dev` we have to mark it as a runtime dependency. + # See discussion on https://github.com/python-poetry/poetry/issues/6154 (it sounds + # like the poetry authors don't consider this a bug?) + # + # In any case, workaround this by ignoring setuptools_rust here. (It might be + # slightly cleaner to put `setuptools_rust` in a `build` extra or similar, but for + # now let's do something quick and dirty. + if req.name == "setuptools_rust": + return True + return False + + class Dependency(NamedTuple): requirement: Requirement must_be_installed: bool @@ -77,7 +92,7 @@ def _generic_dependencies() -> Iterable[Dependency]: assert requirements is not None for raw_requirement in requirements: req = Requirement(raw_requirement) - if _is_dev_dependency(req): + if _is_dev_dependency(req) or _should_ignore_runtime_requirement(req): continue # https://packaging.pypa.io/en/latest/markers.html#usage notes that diff --git a/tests/util/test_check_dependencies.py b/tests/util/test_check_dependencies.py index 5d1aa025d1..6913de24b9 100644 --- a/tests/util/test_check_dependencies.py +++ b/tests/util/test_check_dependencies.py @@ -40,7 +40,10 @@ class TestDependencyChecker(TestCase): def mock_installed_package( self, distribution: Optional[DummyDistribution] ) -> Generator[None, None, None]: - """Pretend that looking up any distribution yields the given `distribution`.""" + """Pretend that looking up any package yields the given `distribution`. + + If `distribution = None`, we pretend that the package is not installed. + """ def mock_distribution(name: str): if distribution is None: @@ -81,7 +84,7 @@ class TestDependencyChecker(TestCase): self.assertRaises(DependencyException, check_requirements) def test_checks_ignore_dev_dependencies(self) -> None: - """Bot generic and per-extra checks should ignore dev dependencies.""" + """Both generic and per-extra checks should ignore dev dependencies.""" with patch( "synapse.util.check_dependencies.metadata.requires", return_value=["dummypkg >= 1; extra == 'mypy'"], @@ -142,3 +145,16 @@ class TestDependencyChecker(TestCase): with self.mock_installed_package(new_release_candidate): # should not raise check_requirements() + + def test_setuptools_rust_ignored(self) -> None: + """Test a workaround for a `poetry build` problem. Reproduces #13926.""" + with patch( + "synapse.util.check_dependencies.metadata.requires", + return_value=["setuptools_rust >= 1.3"], + ): + with self.mock_installed_package(None): + # should not raise, even if setuptools_rust is not installed + check_requirements() + with self.mock_installed_package(old): + # We also ignore old versions of setuptools_rust + check_requirements() -- cgit 1.5.1 From cb20b885cb4bd1648581dd043a184d86fc8c7a00 Mon Sep 17 00:00:00 2001 From: David Robertson Date: Thu, 6 Oct 2022 19:17:50 +0100 Subject: Always close _all_ `ijson` coroutines, even if doing so raises Exceptions (#14065) --- changelog.d/14065.misc | 1 + synapse/federation/transport/client.py | 29 ++++++++++++++++++++---- synapse/util/__init__.py | 14 +++++++++++- tests/federation/transport/test_client.py | 37 +++++++++++++++++++++++++++++++ 4 files changed, 76 insertions(+), 5 deletions(-) create mode 100644 changelog.d/14065.misc (limited to 'synapse/util') diff --git a/changelog.d/14065.misc b/changelog.d/14065.misc new file mode 100644 index 0000000000..98998b0015 --- /dev/null +++ b/changelog.d/14065.misc @@ -0,0 +1 @@ +Fix a bug introduced in Synapse 1.35.0 where errors parsing a `/send_join` or `/state` response would produce excessive, low-quality Sentry events. diff --git a/synapse/federation/transport/client.py b/synapse/federation/transport/client.py index 32074b8ca6..cd39d4d111 100644 --- a/synapse/federation/transport/client.py +++ b/synapse/federation/transport/client.py @@ -45,6 +45,7 @@ from synapse.federation.units import Transaction from synapse.http.matrixfederationclient import ByteParser from synapse.http.types import QueryParams from synapse.types import JsonDict +from synapse.util import ExceptionBundle logger = logging.getLogger(__name__) @@ -926,8 +927,7 @@ class SendJoinParser(ByteParser[SendJoinResponse]): return len(data) def finish(self) -> SendJoinResponse: - for c in self._coros: - c.close() + _close_coros(self._coros) if self._response.event_dict: self._response.event = make_event_from_dict( @@ -970,6 +970,27 @@ class _StateParser(ByteParser[StateRequestResponse]): return len(data) def finish(self) -> StateRequestResponse: - for c in self._coros: - c.close() + _close_coros(self._coros) return self._response + + +def _close_coros(coros: Iterable[Generator[None, bytes, None]]) -> None: + """Close each of the given coroutines. + + Always calls .close() on each coroutine, even if doing so raises an exception. + Any exceptions raised are aggregated into an ExceptionBundle. + + :raises ExceptionBundle: if at least one coroutine fails to close. + """ + exceptions = [] + for c in coros: + try: + c.close() + except Exception as e: + exceptions.append(e) + + if exceptions: + # raise from the first exception so that the traceback has slightly more context + raise ExceptionBundle( + f"There were {len(exceptions)} errors closing coroutines", exceptions + ) from exceptions[0] diff --git a/synapse/util/__init__.py b/synapse/util/__init__.py index a90f08dd4c..7be9d5f113 100644 --- a/synapse/util/__init__.py +++ b/synapse/util/__init__.py @@ -15,7 +15,7 @@ import json import logging import typing -from typing import Any, Callable, Dict, Generator, Optional +from typing import Any, Callable, Dict, Generator, Optional, Sequence import attr from frozendict import frozendict @@ -193,3 +193,15 @@ def log_failure( # Version string with git info. Computed here once so that we don't invoke git multiple # times. SYNAPSE_VERSION = get_distribution_version_string("matrix-synapse", __file__) + + +class ExceptionBundle(Exception): + # A poor stand-in for something like Python 3.11's ExceptionGroup. + # (A backport called `exceptiongroup` exists but seems overkill: we just want a + # container type here.) + def __init__(self, message: str, exceptions: Sequence[Exception]): + parts = [message] + for e in exceptions: + parts.append(str(e)) + super().__init__("\n - ".join(parts)) + self.exceptions = exceptions diff --git a/tests/federation/transport/test_client.py b/tests/federation/transport/test_client.py index c2320ce133..0926e0583d 100644 --- a/tests/federation/transport/test_client.py +++ b/tests/federation/transport/test_client.py @@ -13,6 +13,7 @@ # limitations under the License. import json +from unittest.mock import Mock from synapse.api.room_versions import RoomVersions from synapse.federation.transport.client import SendJoinParser @@ -94,3 +95,39 @@ class SendJoinParserTestCase(TestCase): # Retrieve and check the parsed SendJoinResponse parsed_response = parser.finish() self.assertEqual(parsed_response.servers_in_room, ["hs1", "hs2"]) + + def test_errors_closing_coroutines(self) -> None: + """Check we close all coroutines, even if closing the first raises an Exception. + + We also check that an Exception of some kind is raised, but we don't make any + assertions about its attributes or type. + """ + parser = SendJoinParser(RoomVersions.V1, False) + response = {"org.matrix.msc3706.servers_in_room": ["hs1", "hs2"]} + serialisation = json.dumps(response).encode() + + # Mock the coroutines managed by this parser. + # The first one will error when we try to close it. + coro_1 = Mock() + coro_1.close = Mock(side_effect=RuntimeError("Couldn't close coro 1")) + + coro_2 = Mock() + + coro_3 = Mock() + coro_3.close = Mock(side_effect=RuntimeError("Couldn't close coro 3")) + + parser._coros = [coro_1, coro_2, coro_3] + + # Send half of the data to the parser + parser.write(serialisation[: len(serialisation) // 2]) + + # Close the parser. There should be _some_ kind of exception, but it need not + # be that RuntimeError directly. E.g. we might want to raise a wrapper + # encompassing multiple errors from multiple coroutines. + with self.assertRaises(Exception): + parser.finish() + + # In any case, we should have tried to close both coros. + coro_1.close.assert_called() + coro_2.close.assert_called() + coro_3.close.assert_called() -- cgit 1.5.1 From a9934d48c193bc963e3d232ed83c5cbfa3e5152d Mon Sep 17 00:00:00 2001 From: Abdullah Osama Date: Tue, 11 Oct 2022 14:42:11 +0200 Subject: Making parse_server_name more consistent (#14007) Fixes #12122 --- changelog.d/14007.misc | 1 + synapse/util/stringutils.py | 4 ++-- tests/http/test_endpoint.py | 3 +++ 3 files changed, 6 insertions(+), 2 deletions(-) create mode 100644 changelog.d/14007.misc (limited to 'synapse/util') diff --git a/changelog.d/14007.misc b/changelog.d/14007.misc new file mode 100644 index 0000000000..3f0f3afe1c --- /dev/null +++ b/changelog.d/14007.misc @@ -0,0 +1 @@ +Make `parse_server_name` consistent in handling invalid server names. \ No newline at end of file diff --git a/synapse/util/stringutils.py b/synapse/util/stringutils.py index 27a363d7e5..4961fe9313 100644 --- a/synapse/util/stringutils.py +++ b/synapse/util/stringutils.py @@ -86,7 +86,7 @@ def parse_server_name(server_name: str) -> Tuple[str, Optional[int]]: ValueError if the server name could not be parsed. """ try: - if server_name[-1] == "]": + if server_name and server_name[-1] == "]": # ipv6 literal, hopefully return server_name, None @@ -123,7 +123,7 @@ def parse_and_validate_server_name(server_name: str) -> Tuple[str, Optional[int] # that nobody is sneaking IP literals in that look like hostnames, etc. # look for ipv6 literals - if host[0] == "[": + if host and host[0] == "[": if host[-1] != "]": raise ValueError("Mismatched [...] in server name '%s'" % (server_name,)) diff --git a/tests/http/test_endpoint.py b/tests/http/test_endpoint.py index c8cc21cadd..a801f002a0 100644 --- a/tests/http/test_endpoint.py +++ b/tests/http/test_endpoint.py @@ -25,6 +25,8 @@ class ServerNameTestCase(unittest.TestCase): "[0abc:1def::1234]": ("[0abc:1def::1234]", None), "1.2.3.4:1": ("1.2.3.4", 1), "[0abc:1def::1234]:8080": ("[0abc:1def::1234]", 8080), + ":80": ("", 80), + "": ("", None), } for i, o in test_data.items(): @@ -42,6 +44,7 @@ class ServerNameTestCase(unittest.TestCase): "newline.com\n", ".empty-label.com", "1234:5678:80", # too many colons + ":80", ] for i in test_data: try: -- cgit 1.5.1 From 0b7830e457359ce651b293c8748bf636973404a9 Mon Sep 17 00:00:00 2001 From: "dependabot[bot]" <49699333+dependabot[bot]@users.noreply.github.com> Date: Wed, 19 Oct 2022 19:38:24 +0000 Subject: Bump flake8-bugbear from 21.3.2 to 22.9.23 (#14042) Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com> Co-authored-by: Erik Johnston Co-authored-by: David Robertson --- .flake8 | 9 ++++++++- changelog.d/14042.misc | 1 + poetry.lock | 8 ++++---- synapse/storage/databases/main/roommember.py | 4 ++-- synapse/util/caches/deferred_cache.py | 4 ++-- synapse/util/caches/descriptors.py | 2 +- tests/federation/transport/test_client.py | 7 +++---- tests/util/caches/test_descriptors.py | 2 +- 8 files changed, 22 insertions(+), 15 deletions(-) create mode 100644 changelog.d/14042.misc (limited to 'synapse/util') diff --git a/.flake8 b/.flake8 index acb118c86e..4c6a4d5843 100644 --- a/.flake8 +++ b/.flake8 @@ -8,4 +8,11 @@ # E203: whitespace before ':' (which is contrary to pep8?) # E731: do not assign a lambda expression, use a def # E501: Line too long (black enforces this for us) -ignore=W503,W504,E203,E731,E501 +# +# flake8-bugbear runs extra checks. Its error codes are described at +# https://github.com/PyCQA/flake8-bugbear#list-of-warnings +# B019: Use of functools.lru_cache or functools.cache on methods can lead to memory leaks +# B023: Functions defined inside a loop must not use variables redefined in the loop +# B024: Abstract base class with no abstract method. + +ignore=W503,W504,E203,E731,E501,B019,B023,B024 diff --git a/changelog.d/14042.misc b/changelog.d/14042.misc new file mode 100644 index 0000000000..868d55e76a --- /dev/null +++ b/changelog.d/14042.misc @@ -0,0 +1 @@ +Bump flake8-bugbear from 21.3.2 to 22.9.23. diff --git a/poetry.lock b/poetry.lock index ed0b59fbe5..0a2f9ab69e 100644 --- a/poetry.lock +++ b/poetry.lock @@ -260,7 +260,7 @@ pyflakes = ">=2.4.0,<2.5.0" [[package]] name = "flake8-bugbear" -version = "21.3.2" +version = "22.9.23" description = "A plugin for flake8 finding likely bugs and design problems in your program. Contains warnings that don't belong in pyflakes and pycodestyle." category = "dev" optional = false @@ -271,7 +271,7 @@ attrs = ">=19.2.0" flake8 = ">=3.0.0" [package.extras] -dev = ["black", "coverage", "hypothesis", "hypothesmith"] +dev = ["coverage", "hypothesis", "hypothesmith (>=0.2)", "pre-commit"] [[package]] name = "flake8-comprehensions" @@ -1826,8 +1826,8 @@ flake8 = [ {file = "flake8-4.0.1.tar.gz", hash = "sha256:806e034dda44114815e23c16ef92f95c91e4c71100ff52813adf7132a6ad870d"}, ] flake8-bugbear = [ - {file = "flake8-bugbear-21.3.2.tar.gz", hash = "sha256:cadce434ceef96463b45a7c3000f23527c04ea4b531d16c7ac8886051f516ca0"}, - {file = "flake8_bugbear-21.3.2-py36.py37.py38-none-any.whl", hash = "sha256:5d6ccb0c0676c738a6e066b4d50589c408dcc1c5bf1d73b464b18b73cd6c05c2"}, + {file = "flake8-bugbear-22.9.23.tar.gz", hash = "sha256:17b9623325e6e0dcdcc80ed9e4aa811287fcc81d7e03313b8736ea5733759937"}, + {file = "flake8_bugbear-22.9.23-py3-none-any.whl", hash = "sha256:cd2779b2b7ada212d7a322814a1e5651f1868ab0d3f24cc9da66169ab8fda474"}, ] flake8-comprehensions = [ {file = "flake8-comprehensions-3.8.0.tar.gz", hash = "sha256:8e108707637b1d13734f38e03435984f6b7854fa6b5a4e34f93e69534be8e521"}, diff --git a/synapse/storage/databases/main/roommember.py b/synapse/storage/databases/main/roommember.py index 2ed6ad754f..32e1e983a5 100644 --- a/synapse/storage/databases/main/roommember.py +++ b/synapse/storage/databases/main/roommember.py @@ -707,8 +707,8 @@ class RoomMemberWorkerStore(EventsWorkerStore): # 250 users is pretty arbitrary but the data can be quite large if users # are in many rooms. - for user_ids in batch_iter(user_ids, 250): - all_user_rooms.update(await self._get_rooms_for_users(user_ids)) + for batch_user_ids in batch_iter(user_ids, 250): + all_user_rooms.update(await self._get_rooms_for_users(batch_user_ids)) return all_user_rooms diff --git a/synapse/util/caches/deferred_cache.py b/synapse/util/caches/deferred_cache.py index 6425f851ea..bcb1cba362 100644 --- a/synapse/util/caches/deferred_cache.py +++ b/synapse/util/caches/deferred_cache.py @@ -395,8 +395,8 @@ class DeferredCache(Generic[KT, VT]): # _pending_deferred_cache.pop should either return a CacheEntry, or, in the # case of a TreeCache, a dict of keys to cache entries. Either way calling # iterate_tree_cache_entry on it will do the right thing. - for entry in iterate_tree_cache_entry(entry): - for cb in entry.get_invalidation_callbacks(key): + for iter_entry in iterate_tree_cache_entry(entry): + for cb in iter_entry.get_invalidation_callbacks(key): cb() def invalidate_all(self) -> None: diff --git a/synapse/util/caches/descriptors.py b/synapse/util/caches/descriptors.py index 0391966462..b3c748ef44 100644 --- a/synapse/util/caches/descriptors.py +++ b/synapse/util/caches/descriptors.py @@ -432,7 +432,7 @@ class DeferredCacheListDescriptor(_CacheDescriptorBase): num_args = cached_method.num_args if num_args != self.num_args: - raise Exception( + raise TypeError( "Number of args (%s) does not match underlying cache_method_name=%s (%s)." % (self.num_args, self.cached_method_name, num_args) ) diff --git a/tests/federation/transport/test_client.py b/tests/federation/transport/test_client.py index 0926e0583d..dd4d1b56de 100644 --- a/tests/federation/transport/test_client.py +++ b/tests/federation/transport/test_client.py @@ -17,6 +17,7 @@ from unittest.mock import Mock from synapse.api.room_versions import RoomVersions from synapse.federation.transport.client import SendJoinParser +from synapse.util import ExceptionBundle from tests.unittest import TestCase @@ -121,10 +122,8 @@ class SendJoinParserTestCase(TestCase): # Send half of the data to the parser parser.write(serialisation[: len(serialisation) // 2]) - # Close the parser. There should be _some_ kind of exception, but it need not - # be that RuntimeError directly. E.g. we might want to raise a wrapper - # encompassing multiple errors from multiple coroutines. - with self.assertRaises(Exception): + # Close the parser. There should be _some_ kind of exception. + with self.assertRaises(ExceptionBundle): parser.finish() # In any case, we should have tried to close both coros. diff --git a/tests/util/caches/test_descriptors.py b/tests/util/caches/test_descriptors.py index 90861fe522..78fd7b6961 100644 --- a/tests/util/caches/test_descriptors.py +++ b/tests/util/caches/test_descriptors.py @@ -1037,5 +1037,5 @@ class CachedListDescriptorTestCase(unittest.TestCase): obj = Cls() # Make sure this raises an error about the arg mismatch - with self.assertRaises(Exception): + with self.assertRaises(TypeError): obj.list_fn([("foo", "bar")]) -- cgit 1.5.1 From c9dffd5b330553c5803784be5bc0e2479fab79b0 Mon Sep 17 00:00:00 2001 From: Nick Mills-Barrett Date: Tue, 25 Oct 2022 11:39:25 +0100 Subject: Remove unused `@lru_cache` decorator (#13595) * Remove unused `@lru_cache` decorator Spotted this working on something else. Co-authored-by: David Robertson --- changelog.d/13595.misc | 1 + synapse/util/caches/descriptors.py | 104 ---------------------------------- tests/util/caches/test_descriptors.py | 40 ++----------- 3 files changed, 5 insertions(+), 140 deletions(-) create mode 100644 changelog.d/13595.misc (limited to 'synapse/util') diff --git a/changelog.d/13595.misc b/changelog.d/13595.misc new file mode 100644 index 0000000000..71959a6ee7 --- /dev/null +++ b/changelog.d/13595.misc @@ -0,0 +1 @@ +Remove unused `@lru_cache` decorator. diff --git a/synapse/util/caches/descriptors.py b/synapse/util/caches/descriptors.py index b3c748ef44..75428d19ba 100644 --- a/synapse/util/caches/descriptors.py +++ b/synapse/util/caches/descriptors.py @@ -12,7 +12,6 @@ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. -import enum import functools import inspect import logging @@ -146,109 +145,6 @@ class _CacheDescriptorBase: ) -class _LruCachedFunction(Generic[F]): - cache: LruCache[CacheKey, Any] - __call__: F - - -def lru_cache( - *, max_entries: int = 1000, cache_context: bool = False -) -> Callable[[F], _LruCachedFunction[F]]: - """A method decorator that applies a memoizing cache around the function. - - This is more-or-less a drop-in equivalent to functools.lru_cache, although note - that the signature is slightly different. - - The main differences with functools.lru_cache are: - (a) the size of the cache can be controlled via the cache_factor mechanism - (b) the wrapped function can request a "cache_context" which provides a - callback mechanism to indicate that the result is no longer valid - (c) prometheus metrics are exposed automatically. - - The function should take zero or more arguments, which are used as the key for the - cache. Single-argument functions use that argument as the cache key; otherwise the - arguments are built into a tuple. - - Cached functions can be "chained" (i.e. a cached function can call other cached - functions and get appropriately invalidated when they called caches are - invalidated) by adding a special "cache_context" argument to the function - and passing that as a kwarg to all caches called. For example: - - @lru_cache(cache_context=True) - def foo(self, key, cache_context): - r1 = self.bar1(key, on_invalidate=cache_context.invalidate) - r2 = self.bar2(key, on_invalidate=cache_context.invalidate) - return r1 + r2 - - The wrapped function also has a 'cache' property which offers direct access to the - underlying LruCache. - """ - - def func(orig: F) -> _LruCachedFunction[F]: - desc = LruCacheDescriptor( - orig, - max_entries=max_entries, - cache_context=cache_context, - ) - return cast(_LruCachedFunction[F], desc) - - return func - - -class LruCacheDescriptor(_CacheDescriptorBase): - """Helper for @lru_cache""" - - class _Sentinel(enum.Enum): - sentinel = object() - - def __init__( - self, - orig: Callable[..., Any], - max_entries: int = 1000, - cache_context: bool = False, - ): - super().__init__( - orig, num_args=None, uncached_args=None, cache_context=cache_context - ) - self.max_entries = max_entries - - def __get__(self, obj: Optional[Any], owner: Optional[Type]) -> Callable[..., Any]: - cache: LruCache[CacheKey, Any] = LruCache( - cache_name=self.name, - max_size=self.max_entries, - ) - - get_cache_key = self.cache_key_builder - sentinel = LruCacheDescriptor._Sentinel.sentinel - - @functools.wraps(self.orig) - def _wrapped(*args: Any, **kwargs: Any) -> Any: - invalidate_callback = kwargs.pop("on_invalidate", None) - callbacks = (invalidate_callback,) if invalidate_callback else () - - cache_key = get_cache_key(args, kwargs) - - ret = cache.get(cache_key, default=sentinel, callbacks=callbacks) - if ret != sentinel: - return ret - - # Add our own `cache_context` to argument list if the wrapped function - # has asked for one - if self.add_cache_context: - kwargs["cache_context"] = _CacheContext.get_instance(cache, cache_key) - - ret2 = self.orig(obj, *args, **kwargs) - cache.set(cache_key, ret2, callbacks=callbacks) - - return ret2 - - wrapped = cast(CachedFunction, _wrapped) - wrapped.cache = cache - obj.__dict__[self.name] = wrapped - - return wrapped - - class DeferredCacheDescriptor(_CacheDescriptorBase): """A method decorator that applies a memoizing cache around the function. diff --git a/tests/util/caches/test_descriptors.py b/tests/util/caches/test_descriptors.py index 78fd7b6961..43475a307f 100644 --- a/tests/util/caches/test_descriptors.py +++ b/tests/util/caches/test_descriptors.py @@ -28,7 +28,7 @@ from synapse.logging.context import ( make_deferred_yieldable, ) from synapse.util.caches import descriptors -from synapse.util.caches.descriptors import cached, cachedList, lru_cache +from synapse.util.caches.descriptors import cached, cachedList from tests import unittest from tests.test_utils import get_awaitable_result @@ -36,38 +36,6 @@ from tests.test_utils import get_awaitable_result logger = logging.getLogger(__name__) -class LruCacheDecoratorTestCase(unittest.TestCase): - def test_base(self): - class Cls: - def __init__(self): - self.mock = mock.Mock() - - @lru_cache() - def fn(self, arg1, arg2): - return self.mock(arg1, arg2) - - obj = Cls() - obj.mock.return_value = "fish" - r = obj.fn(1, 2) - self.assertEqual(r, "fish") - obj.mock.assert_called_once_with(1, 2) - obj.mock.reset_mock() - - # a call with different params should call the mock again - obj.mock.return_value = "chips" - r = obj.fn(1, 3) - self.assertEqual(r, "chips") - obj.mock.assert_called_once_with(1, 3) - obj.mock.reset_mock() - - # the two values should now be cached - r = obj.fn(1, 2) - self.assertEqual(r, "fish") - r = obj.fn(1, 3) - self.assertEqual(r, "chips") - obj.mock.assert_not_called() - - def run_on_reactor(): d = defer.Deferred() reactor.callLater(0, d.callback, 0) @@ -478,10 +446,10 @@ class DescriptorTestCase(unittest.TestCase): @cached(cache_context=True) async def func2(self, key, cache_context): - return self.func3(key, on_invalidate=cache_context.invalidate) + return await self.func3(key, on_invalidate=cache_context.invalidate) - @lru_cache(cache_context=True) - def func3(self, key, cache_context): + @cached(cache_context=True) + async def func3(self, key, cache_context): self.invalidate = cache_context.invalidate return 42 -- cgit 1.5.1 From 8756d5c87efc5637da55c9e21d2a4eb2369ba693 Mon Sep 17 00:00:00 2001 From: Quentin Gliech Date: Wed, 26 Oct 2022 12:45:41 +0200 Subject: Save login tokens in database (#13844) * Save login tokens in database Signed-off-by: Quentin Gliech * Add upgrade notes * Track login token reuse in a Prometheus metric Signed-off-by: Quentin Gliech --- changelog.d/13844.misc | 1 + docs/upgrade.md | 9 ++ synapse/handlers/auth.py | 64 +++++++-- synapse/module_api/__init__.py | 41 +----- synapse/rest/client/login.py | 3 +- synapse/rest/client/login_token_request.py | 5 +- synapse/storage/databases/main/registration.py | 156 ++++++++++++++++++++- .../schema/main/delta/73/10login_tokens.sql | 35 +++++ synapse/util/macaroons.py | 87 +----------- tests/handlers/test_auth.py | 135 ++++++++++-------- tests/util/test_macaroons.py | 28 ---- 11 files changed, 337 insertions(+), 227 deletions(-) create mode 100644 changelog.d/13844.misc create mode 100644 synapse/storage/schema/main/delta/73/10login_tokens.sql (limited to 'synapse/util') diff --git a/changelog.d/13844.misc b/changelog.d/13844.misc new file mode 100644 index 0000000000..66f4414df7 --- /dev/null +++ b/changelog.d/13844.misc @@ -0,0 +1 @@ +Save login tokens in database and prevent login token reuse. diff --git a/docs/upgrade.md b/docs/upgrade.md index b81385b191..78c34d0c15 100644 --- a/docs/upgrade.md +++ b/docs/upgrade.md @@ -88,6 +88,15 @@ process, for example: dpkg -i matrix-synapse-py3_1.3.0+stretch1_amd64.deb ``` +# Upgrading to v1.71.0 + +## Removal of the `generate_short_term_login_token` module API method + +As announced with the release of [Synapse 1.69.0](#deprecation-of-the-generate_short_term_login_token-module-api-method), the deprecated `generate_short_term_login_token` module method has been removed. + +Modules relying on it can instead use the `create_login_token` method. + + # Upgrading to v1.69.0 ## Changes to the receipts replication streams diff --git a/synapse/handlers/auth.py b/synapse/handlers/auth.py index f5f0e0e7a7..8b9ef25d29 100644 --- a/synapse/handlers/auth.py +++ b/synapse/handlers/auth.py @@ -38,6 +38,7 @@ from typing import ( import attr import bcrypt import unpaddedbase64 +from prometheus_client import Counter from twisted.internet.defer import CancelledError from twisted.web.server import Request @@ -48,6 +49,7 @@ from synapse.api.errors import ( Codes, InteractiveAuthIncompleteError, LoginError, + NotFoundError, StoreError, SynapseError, UserDeactivatedError, @@ -63,10 +65,14 @@ from synapse.http.server import finish_request, respond_with_html from synapse.http.site import SynapseRequest from synapse.logging.context import defer_to_thread from synapse.metrics.background_process_metrics import run_as_background_process +from synapse.storage.databases.main.registration import ( + LoginTokenExpired, + LoginTokenLookupResult, + LoginTokenReused, +) from synapse.types import JsonDict, Requester, UserID from synapse.util import stringutils as stringutils from synapse.util.async_helpers import delay_cancellation, maybe_awaitable -from synapse.util.macaroons import LoginTokenAttributes from synapse.util.msisdn import phone_number_to_msisdn from synapse.util.stringutils import base62_encode from synapse.util.threepids import canonicalise_email @@ -80,6 +86,12 @@ logger = logging.getLogger(__name__) INVALID_USERNAME_OR_PASSWORD = "Invalid username or password" +invalid_login_token_counter = Counter( + "synapse_user_login_invalid_login_tokens", + "Counts the number of rejected m.login.token on /login", + ["reason"], +) + def convert_client_dict_legacy_fields_to_identifier( submission: JsonDict, @@ -883,6 +895,25 @@ class AuthHandler: return True + async def create_login_token_for_user_id( + self, + user_id: str, + duration_ms: int = (2 * 60 * 1000), + auth_provider_id: Optional[str] = None, + auth_provider_session_id: Optional[str] = None, + ) -> str: + login_token = self.generate_login_token() + now = self._clock.time_msec() + expiry_ts = now + duration_ms + await self.store.add_login_token_to_user( + user_id=user_id, + token=login_token, + expiry_ts=expiry_ts, + auth_provider_id=auth_provider_id, + auth_provider_session_id=auth_provider_session_id, + ) + return login_token + async def create_refresh_token_for_user_id( self, user_id: str, @@ -1401,6 +1432,18 @@ class AuthHandler: return None return user_id + def generate_login_token(self) -> str: + """Generates an opaque string, for use as an short-term login token""" + + # we use the following format for access tokens: + # syl__ + + random_string = stringutils.random_string(20) + base = f"syl_{random_string}" + + crc = base62_encode(crc32(base.encode("ascii")), minwidth=6) + return f"{base}_{crc}" + def generate_access_token(self, for_user: UserID) -> str: """Generates an opaque string, for use as an access token""" @@ -1427,16 +1470,17 @@ class AuthHandler: crc = base62_encode(crc32(base.encode("ascii")), minwidth=6) return f"{base}_{crc}" - async def validate_short_term_login_token( - self, login_token: str - ) -> LoginTokenAttributes: + async def consume_login_token(self, login_token: str) -> LoginTokenLookupResult: try: - res = self.macaroon_gen.verify_short_term_login_token(login_token) - except Exception: - raise AuthError(403, "Invalid login token", errcode=Codes.FORBIDDEN) + return await self.store.consume_login_token(login_token) + except LoginTokenExpired: + invalid_login_token_counter.labels("expired").inc() + except LoginTokenReused: + invalid_login_token_counter.labels("reused").inc() + except NotFoundError: + invalid_login_token_counter.labels("not found").inc() - await self.auth_blocking.check_auth_blocking(res.user_id) - return res + raise AuthError(403, "Invalid login token", errcode=Codes.FORBIDDEN) async def delete_access_token(self, access_token: str) -> None: """Invalidate a single access token @@ -1711,7 +1755,7 @@ class AuthHandler: ) # Create a login token - login_token = self.macaroon_gen.generate_short_term_login_token( + login_token = await self.create_login_token_for_user_id( registered_user_id, auth_provider_id=auth_provider_id, auth_provider_session_id=auth_provider_session_id, diff --git a/synapse/module_api/__init__.py b/synapse/module_api/__init__.py index 6a6ae208d1..30e689d00d 100644 --- a/synapse/module_api/__init__.py +++ b/synapse/module_api/__init__.py @@ -771,50 +771,11 @@ class ModuleApi: auth_provider_session_id: The session ID got during login from the SSO IdP, if any. """ - # The deprecated `generate_short_term_login_token` method defaulted to an empty - # string for the `auth_provider_id` because of how the underlying macaroon was - # generated. This will change to a proper NULL-able field when the tokens get - # moved to the database. - return self._hs.get_macaroon_generator().generate_short_term_login_token( + return await self._hs.get_auth_handler().create_login_token_for_user_id( user_id, - auth_provider_id or "", - auth_provider_session_id, duration_in_ms, - ) - - def generate_short_term_login_token( - self, - user_id: str, - duration_in_ms: int = (2 * 60 * 1000), - auth_provider_id: str = "", - auth_provider_session_id: Optional[str] = None, - ) -> str: - """Generate a login token suitable for m.login.token authentication - - Added in Synapse v1.9.0. - - This was deprecated in Synapse v1.69.0 in favor of create_login_token, and will - be removed in Synapse 1.71.0. - - Args: - user_id: gives the ID of the user that the token is for - - duration_in_ms: the time that the token will be valid for - - auth_provider_id: the ID of the SSO IdP that the user used to authenticate - to get this token, if any. This is encoded in the token so that - /login can report stats on number of successful logins by IdP. - """ - logger.warn( - "A module configured on this server uses ModuleApi.generate_short_term_login_token(), " - "which is deprecated in favor of ModuleApi.create_login_token(), and will be removed in " - "Synapse 1.71.0", - ) - return self._hs.get_macaroon_generator().generate_short_term_login_token( - user_id, auth_provider_id, auth_provider_session_id, - duration_in_ms, ) @defer.inlineCallbacks diff --git a/synapse/rest/client/login.py b/synapse/rest/client/login.py index f554586ac3..7774f1967d 100644 --- a/synapse/rest/client/login.py +++ b/synapse/rest/client/login.py @@ -436,8 +436,7 @@ class LoginRestServlet(RestServlet): The body of the JSON response. """ token = login_submission["token"] - auth_handler = self.auth_handler - res = await auth_handler.validate_short_term_login_token(token) + res = await self.auth_handler.consume_login_token(token) return await self._complete_login( res.user_id, diff --git a/synapse/rest/client/login_token_request.py b/synapse/rest/client/login_token_request.py index 277b20fb63..43ea21d5e6 100644 --- a/synapse/rest/client/login_token_request.py +++ b/synapse/rest/client/login_token_request.py @@ -57,7 +57,6 @@ class LoginTokenRequestServlet(RestServlet): self.store = hs.get_datastores().main self.clock = hs.get_clock() self.server_name = hs.config.server.server_name - self.macaroon_gen = hs.get_macaroon_generator() self.auth_handler = hs.get_auth_handler() self.token_timeout = hs.config.experimental.msc3882_token_timeout self.ui_auth = hs.config.experimental.msc3882_ui_auth @@ -76,10 +75,10 @@ class LoginTokenRequestServlet(RestServlet): can_skip_ui_auth=False, # Don't allow skipping of UI auth ) - login_token = self.macaroon_gen.generate_short_term_login_token( + login_token = await self.auth_handler.create_login_token_for_user_id( user_id=requester.user.to_string(), auth_provider_id="org.matrix.msc3882.login_token_request", - duration_in_ms=self.token_timeout, + duration_ms=self.token_timeout, ) return ( diff --git a/synapse/storage/databases/main/registration.py b/synapse/storage/databases/main/registration.py index 2996d6bb4d..0255295317 100644 --- a/synapse/storage/databases/main/registration.py +++ b/synapse/storage/databases/main/registration.py @@ -21,7 +21,13 @@ from typing import TYPE_CHECKING, Any, Dict, List, Optional, Tuple, Union, cast import attr from synapse.api.constants import UserTypes -from synapse.api.errors import Codes, StoreError, SynapseError, ThreepidValidationError +from synapse.api.errors import ( + Codes, + NotFoundError, + StoreError, + SynapseError, + ThreepidValidationError, +) from synapse.config.homeserver import HomeServerConfig from synapse.metrics.background_process_metrics import wrap_as_background_process from synapse.storage.database import ( @@ -50,6 +56,14 @@ class ExternalIDReuseException(Exception): because this external id is given to an other user.""" +class LoginTokenExpired(Exception): + """Exception if the login token sent expired""" + + +class LoginTokenReused(Exception): + """Exception if the login token sent was already used""" + + @attr.s(frozen=True, slots=True, auto_attribs=True) class TokenLookupResult: """Result of looking up an access token. @@ -115,6 +129,20 @@ class RefreshTokenLookupResult: If None, the session can be refreshed indefinitely.""" +@attr.s(auto_attribs=True, frozen=True, slots=True) +class LoginTokenLookupResult: + """Result of looking up a login token.""" + + user_id: str + """The user this token belongs to.""" + + auth_provider_id: Optional[str] + """The SSO Identity Provider that the user authenticated with, to get this token.""" + + auth_provider_session_id: Optional[str] + """The session ID advertised by the SSO Identity Provider.""" + + class RegistrationWorkerStore(CacheInvalidationWorkerStore): def __init__( self, @@ -1789,6 +1817,109 @@ class RegistrationWorkerStore(CacheInvalidationWorkerStore): "replace_refresh_token", _replace_refresh_token_txn ) + async def add_login_token_to_user( + self, + user_id: str, + token: str, + expiry_ts: int, + auth_provider_id: Optional[str], + auth_provider_session_id: Optional[str], + ) -> None: + """Adds a short-term login token for the given user. + + Args: + user_id: The user ID. + token: The new login token to add. + expiry_ts (milliseconds since the epoch): Time after which the login token + cannot be used. + auth_provider_id: The SSO Identity Provider that the user authenticated with + to get this token, if any + auth_provider_session_id: The session ID advertised by the SSO Identity + Provider, if any. + """ + await self.db_pool.simple_insert( + "login_tokens", + { + "token": token, + "user_id": user_id, + "expiry_ts": expiry_ts, + "auth_provider_id": auth_provider_id, + "auth_provider_session_id": auth_provider_session_id, + }, + desc="add_login_token_to_user", + ) + + def _consume_login_token( + self, + txn: LoggingTransaction, + token: str, + ts: int, + ) -> LoginTokenLookupResult: + values = self.db_pool.simple_select_one_txn( + txn, + "login_tokens", + keyvalues={"token": token}, + retcols=( + "user_id", + "expiry_ts", + "used_ts", + "auth_provider_id", + "auth_provider_session_id", + ), + allow_none=True, + ) + + if values is None: + raise NotFoundError() + + self.db_pool.simple_update_one_txn( + txn, + "login_tokens", + keyvalues={"token": token}, + updatevalues={"used_ts": ts}, + ) + user_id = values["user_id"] + expiry_ts = values["expiry_ts"] + used_ts = values["used_ts"] + auth_provider_id = values["auth_provider_id"] + auth_provider_session_id = values["auth_provider_session_id"] + + # Token was already used + if used_ts is not None: + raise LoginTokenReused() + + # Token expired + if ts > int(expiry_ts): + raise LoginTokenExpired() + + return LoginTokenLookupResult( + user_id=user_id, + auth_provider_id=auth_provider_id, + auth_provider_session_id=auth_provider_session_id, + ) + + async def consume_login_token(self, token: str) -> LoginTokenLookupResult: + """Lookup a login token and consume it. + + Args: + token: The login token. + + Returns: + The data stored with that token, including the `user_id`. Returns `None` if + the token does not exist or if it expired. + + Raises: + NotFound if the login token was not found in database + LoginTokenExpired if the login token expired + LoginTokenReused if the login token was already used + """ + return await self.db_pool.runInteraction( + "consume_login_token", + self._consume_login_token, + token, + self._clock.time_msec(), + ) + @cached() async def is_guest(self, user_id: str) -> bool: res = await self.db_pool.simple_select_one_onecol( @@ -2019,6 +2150,12 @@ class RegistrationStore(StatsStore, RegistrationBackgroundUpdateStore): and hs.config.experimental.msc3866.require_approval_for_new_accounts ) + # Create a background job for removing expired login tokens + if hs.config.worker.run_background_tasks: + self._clock.looping_call( + self._delete_expired_login_tokens, THIRTY_MINUTES_IN_MS + ) + async def add_access_token_to_user( self, user_id: str, @@ -2617,6 +2754,23 @@ class RegistrationStore(StatsStore, RegistrationBackgroundUpdateStore): approved, ) + @wrap_as_background_process("delete_expired_login_tokens") + async def _delete_expired_login_tokens(self) -> None: + """Remove login tokens with expiry dates that have passed.""" + + def _delete_expired_login_tokens_txn(txn: LoggingTransaction, ts: int) -> None: + sql = "DELETE FROM login_tokens WHERE expiry_ts <= ?" + txn.execute(sql, (ts,)) + + # We keep the expired tokens for an extra 5 minutes so we can measure how many + # times a token is being used after its expiry + now = self._clock.time_msec() + await self.db_pool.runInteraction( + "delete_expired_login_tokens", + _delete_expired_login_tokens_txn, + now - (5 * 60 * 1000), + ) + def find_max_generated_user_id_localpart(cur: Cursor) -> int: """ diff --git a/synapse/storage/schema/main/delta/73/10login_tokens.sql b/synapse/storage/schema/main/delta/73/10login_tokens.sql new file mode 100644 index 0000000000..a39b7bcece --- /dev/null +++ b/synapse/storage/schema/main/delta/73/10login_tokens.sql @@ -0,0 +1,35 @@ +/* + * Copyright 2022 The Matrix.org Foundation C.I.C. + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +-- Login tokens are short-lived tokens that are used for the m.login.token +-- login method, mainly during SSO logins +CREATE TABLE login_tokens ( + token TEXT PRIMARY KEY, + user_id TEXT NOT NULL, + expiry_ts BIGINT NOT NULL, + used_ts BIGINT, + auth_provider_id TEXT, + auth_provider_session_id TEXT +); + +-- We're sometimes querying them by their session ID we got from their IDP +CREATE INDEX login_tokens_auth_provider_idx + ON login_tokens (auth_provider_id, auth_provider_session_id); + +-- We're deleting them by their expiration time +CREATE INDEX login_tokens_expiry_time_idx + ON login_tokens (expiry_ts); + diff --git a/synapse/util/macaroons.py b/synapse/util/macaroons.py index df77edcce2..5df03d3ddc 100644 --- a/synapse/util/macaroons.py +++ b/synapse/util/macaroons.py @@ -24,7 +24,7 @@ from typing_extensions import Literal from synapse.util import Clock, stringutils -MacaroonType = Literal["access", "delete_pusher", "session", "login"] +MacaroonType = Literal["access", "delete_pusher", "session"] def get_value_from_macaroon(macaroon: pymacaroons.Macaroon, key: str) -> str: @@ -111,19 +111,6 @@ class OidcSessionData: """The session ID of the ongoing UI Auth ("" if this is a login)""" -@attr.s(slots=True, frozen=True, auto_attribs=True) -class LoginTokenAttributes: - """Data we store in a short-term login token""" - - user_id: str - - auth_provider_id: str - """The SSO Identity Provider that the user authenticated with, to get this token.""" - - auth_provider_session_id: Optional[str] - """The session ID advertised by the SSO Identity Provider.""" - - class MacaroonGenerator: def __init__(self, clock: Clock, location: str, secret_key: bytes): self._clock = clock @@ -165,35 +152,6 @@ class MacaroonGenerator: macaroon.add_first_party_caveat(f"pushkey = {pushkey}") return macaroon.serialize() - def generate_short_term_login_token( - self, - user_id: str, - auth_provider_id: str, - auth_provider_session_id: Optional[str] = None, - duration_in_ms: int = (2 * 60 * 1000), - ) -> str: - """Generate a short-term login token used during SSO logins - - Args: - user_id: The user for which the token is valid. - auth_provider_id: The SSO IdP the user used. - auth_provider_session_id: The session ID got during login from the SSO IdP. - - Returns: - A signed token valid for using as a ``m.login.token`` token. - """ - now = self._clock.time_msec() - expiry = now + duration_in_ms - macaroon = self._generate_base_macaroon("login") - macaroon.add_first_party_caveat(f"user_id = {user_id}") - macaroon.add_first_party_caveat(f"time < {expiry}") - macaroon.add_first_party_caveat(f"auth_provider_id = {auth_provider_id}") - if auth_provider_session_id is not None: - macaroon.add_first_party_caveat( - f"auth_provider_session_id = {auth_provider_session_id}" - ) - return macaroon.serialize() - def generate_oidc_session_token( self, state: str, @@ -233,49 +191,6 @@ class MacaroonGenerator: return macaroon.serialize() - def verify_short_term_login_token(self, token: str) -> LoginTokenAttributes: - """Verify a short-term-login macaroon - - Checks that the given token is a valid, unexpired short-term-login token - minted by this server. - - Args: - token: The login token to verify. - - Returns: - A set of attributes carried by this token, including the - ``user_id`` and informations about the SSO IDP used during that - login. - - Raises: - MacaroonVerificationFailedException if the verification failed - """ - macaroon = pymacaroons.Macaroon.deserialize(token) - - v = self._base_verifier("login") - v.satisfy_general(lambda c: c.startswith("user_id = ")) - v.satisfy_general(lambda c: c.startswith("auth_provider_id = ")) - v.satisfy_general(lambda c: c.startswith("auth_provider_session_id = ")) - satisfy_expiry(v, self._clock.time_msec) - v.verify(macaroon, self._secret_key) - - user_id = get_value_from_macaroon(macaroon, "user_id") - auth_provider_id = get_value_from_macaroon(macaroon, "auth_provider_id") - - auth_provider_session_id: Optional[str] = None - try: - auth_provider_session_id = get_value_from_macaroon( - macaroon, "auth_provider_session_id" - ) - except MacaroonVerificationFailedException: - pass - - return LoginTokenAttributes( - user_id=user_id, - auth_provider_id=auth_provider_id, - auth_provider_session_id=auth_provider_session_id, - ) - def verify_guest_token(self, token: str) -> str: """Verify a guest access token macaroon diff --git a/tests/handlers/test_auth.py b/tests/handlers/test_auth.py index 7106799d44..036dbbc45b 100644 --- a/tests/handlers/test_auth.py +++ b/tests/handlers/test_auth.py @@ -11,6 +11,7 @@ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. +from typing import Optional from unittest.mock import Mock import pymacaroons @@ -19,6 +20,7 @@ from twisted.test.proto_helpers import MemoryReactor from synapse.api.errors import AuthError, ResourceLimitError from synapse.rest import admin +from synapse.rest.client import login from synapse.server import HomeServer from synapse.util import Clock @@ -29,6 +31,7 @@ from tests.test_utils import make_awaitable class AuthTestCase(unittest.HomeserverTestCase): servlets = [ admin.register_servlets, + login.register_servlets, ] def prepare(self, reactor: MemoryReactor, clock: Clock, hs: HomeServer) -> None: @@ -46,6 +49,23 @@ class AuthTestCase(unittest.HomeserverTestCase): self.user1 = self.register_user("a_user", "pass") + def token_login(self, token: str) -> Optional[str]: + body = { + "type": "m.login.token", + "token": token, + } + + channel = self.make_request( + "POST", + "/_matrix/client/v3/login", + body, + ) + + if channel.code == 200: + return channel.json_body["user_id"] + + return None + def test_macaroon_caveats(self) -> None: token = self.macaroon_generator.generate_guest_access_token("a_user") macaroon = pymacaroons.Macaroon.deserialize(token) @@ -73,49 +93,62 @@ class AuthTestCase(unittest.HomeserverTestCase): v.satisfy_general(verify_guest) v.verify(macaroon, self.hs.config.key.macaroon_secret_key) - def test_short_term_login_token_gives_user_id(self) -> None: - token = self.macaroon_generator.generate_short_term_login_token( - self.user1, "", duration_in_ms=5000 + def test_login_token_gives_user_id(self) -> None: + token = self.get_success( + self.auth_handler.create_login_token_for_user_id( + self.user1, + duration_ms=(5 * 1000), + ) ) - res = self.get_success(self.auth_handler.validate_short_term_login_token(token)) + + res = self.get_success(self.auth_handler.consume_login_token(token)) self.assertEqual(self.user1, res.user_id) - self.assertEqual("", res.auth_provider_id) + self.assertEqual(None, res.auth_provider_id) - # when we advance the clock, the token should be rejected - self.reactor.advance(6) - self.get_failure( - self.auth_handler.validate_short_term_login_token(token), - AuthError, + def test_login_token_reuse_fails(self) -> None: + token = self.get_success( + self.auth_handler.create_login_token_for_user_id( + self.user1, + duration_ms=(5 * 1000), + ) ) - def test_short_term_login_token_gives_auth_provider(self) -> None: - token = self.macaroon_generator.generate_short_term_login_token( - self.user1, auth_provider_id="my_idp" - ) - res = self.get_success(self.auth_handler.validate_short_term_login_token(token)) - self.assertEqual(self.user1, res.user_id) - self.assertEqual("my_idp", res.auth_provider_id) + self.get_success(self.auth_handler.consume_login_token(token)) - def test_short_term_login_token_cannot_replace_user_id(self) -> None: - token = self.macaroon_generator.generate_short_term_login_token( - self.user1, "", duration_in_ms=5000 + self.get_failure( + self.auth_handler.consume_login_token(token), + AuthError, ) - macaroon = pymacaroons.Macaroon.deserialize(token) - res = self.get_success( - self.auth_handler.validate_short_term_login_token(macaroon.serialize()) + def test_login_token_expires(self) -> None: + token = self.get_success( + self.auth_handler.create_login_token_for_user_id( + self.user1, + duration_ms=(5 * 1000), + ) ) - self.assertEqual(self.user1, res.user_id) - - # add another "user_id" caveat, which might allow us to override the - # user_id. - macaroon.add_first_party_caveat("user_id = b_user") + # when we advance the clock, the token should be rejected + self.reactor.advance(6) self.get_failure( - self.auth_handler.validate_short_term_login_token(macaroon.serialize()), + self.auth_handler.consume_login_token(token), AuthError, ) + def test_login_token_gives_auth_provider(self) -> None: + token = self.get_success( + self.auth_handler.create_login_token_for_user_id( + self.user1, + auth_provider_id="my_idp", + auth_provider_session_id="11-22-33-44", + duration_ms=(5 * 1000), + ) + ) + res = self.get_success(self.auth_handler.consume_login_token(token)) + self.assertEqual(self.user1, res.user_id) + self.assertEqual("my_idp", res.auth_provider_id) + self.assertEqual("11-22-33-44", res.auth_provider_session_id) + def test_mau_limits_disabled(self) -> None: self.auth_blocking._limit_usage_by_mau = False # Ensure does not throw exception @@ -125,12 +158,12 @@ class AuthTestCase(unittest.HomeserverTestCase): ) ) - self.get_success( - self.auth_handler.validate_short_term_login_token( - self._get_macaroon().serialize() - ) + token = self.get_success( + self.auth_handler.create_login_token_for_user_id(self.user1) ) + self.assertIsNotNone(self.token_login(token)) + def test_mau_limits_exceeded_large(self) -> None: self.auth_blocking._limit_usage_by_mau = True self.hs.get_datastores().main.get_monthly_active_count = Mock( @@ -147,12 +180,10 @@ class AuthTestCase(unittest.HomeserverTestCase): self.hs.get_datastores().main.get_monthly_active_count = Mock( return_value=make_awaitable(self.large_number_of_users) ) - self.get_failure( - self.auth_handler.validate_short_term_login_token( - self._get_macaroon().serialize() - ), - ResourceLimitError, + token = self.get_success( + self.auth_handler.create_login_token_for_user_id(self.user1) ) + self.assertIsNone(self.token_login(token)) def test_mau_limits_parity(self) -> None: # Ensure we're not at the unix epoch. @@ -171,12 +202,10 @@ class AuthTestCase(unittest.HomeserverTestCase): ), ResourceLimitError, ) - self.get_failure( - self.auth_handler.validate_short_term_login_token( - self._get_macaroon().serialize() - ), - ResourceLimitError, + token = self.get_success( + self.auth_handler.create_login_token_for_user_id(self.user1) ) + self.assertIsNone(self.token_login(token)) # If in monthly active cohort self.hs.get_datastores().main.user_last_seen_monthly_active = Mock( @@ -187,11 +216,10 @@ class AuthTestCase(unittest.HomeserverTestCase): self.user1, device_id=None, valid_until_ms=None ) ) - self.get_success( - self.auth_handler.validate_short_term_login_token( - self._get_macaroon().serialize() - ) + token = self.get_success( + self.auth_handler.create_login_token_for_user_id(self.user1) ) + self.assertIsNotNone(self.token_login(token)) def test_mau_limits_not_exceeded(self) -> None: self.auth_blocking._limit_usage_by_mau = True @@ -209,14 +237,7 @@ class AuthTestCase(unittest.HomeserverTestCase): self.hs.get_datastores().main.get_monthly_active_count = Mock( return_value=make_awaitable(self.small_number_of_users) ) - self.get_success( - self.auth_handler.validate_short_term_login_token( - self._get_macaroon().serialize() - ) - ) - - def _get_macaroon(self) -> pymacaroons.Macaroon: - token = self.macaroon_generator.generate_short_term_login_token( - self.user1, "", duration_in_ms=5000 + token = self.get_success( + self.auth_handler.create_login_token_for_user_id(self.user1) ) - return pymacaroons.Macaroon.deserialize(token) + self.assertIsNotNone(self.token_login(token)) diff --git a/tests/util/test_macaroons.py b/tests/util/test_macaroons.py index 32125f7bb7..40754a4711 100644 --- a/tests/util/test_macaroons.py +++ b/tests/util/test_macaroons.py @@ -84,34 +84,6 @@ class MacaroonGeneratorTestCase(TestCase): ) self.assertEqual(user_id, "@user:tesths") - def test_short_term_login_token(self): - """Test the generation and verification of short-term login tokens""" - token = self.macaroon_generator.generate_short_term_login_token( - user_id="@user:tesths", - auth_provider_id="oidc", - auth_provider_session_id="sid", - duration_in_ms=2 * 60 * 1000, - ) - - info = self.macaroon_generator.verify_short_term_login_token(token) - self.assertEqual(info.user_id, "@user:tesths") - self.assertEqual(info.auth_provider_id, "oidc") - self.assertEqual(info.auth_provider_session_id, "sid") - - # Raises with another secret key - with self.assertRaises(MacaroonVerificationFailedException): - self.other_macaroon_generator.verify_short_term_login_token(token) - - # Wait a minute - self.reactor.pump([60]) - # Shouldn't raise - self.macaroon_generator.verify_short_term_login_token(token) - # Wait another minute - self.reactor.pump([60]) - # Should raise since it expired - with self.assertRaises(MacaroonVerificationFailedException): - self.macaroon_generator.verify_short_term_login_token(token) - def test_oidc_session_token(self): """Test the generation and verification of OIDC session cookies""" state = "arandomstate" -- cgit 1.5.1 From 40fa8294e3096132819287dd0c6d6bd71a408902 Mon Sep 17 00:00:00 2001 From: Eric Eastwood Date: Wed, 26 Oct 2022 16:10:55 -0500 Subject: Refactor MSC3030 `/timestamp_to_event` to move away from our snowflake pull from `destination` pattern (#14096) 1. `federation_client.timestamp_to_event(...)` now handles all `destination` looping and uses our generic `_try_destination_list(...)` helper. 2. Consistently handling `NotRetryingDestination` and `FederationDeniedError` across `get_pdu` , backfill, and the generic `_try_destination_list` which is used for many places we use this pattern. 3. `get_pdu(...)` now returns `PulledPduInfo` so we know which `destination` we ended up pulling the PDU from --- changelog.d/14096.misc | 1 + synapse/federation/federation_client.py | 130 ++++++++++++++++++++++++----- synapse/handlers/federation.py | 15 ++-- synapse/handlers/federation_event.py | 31 ++++--- synapse/handlers/room.py | 126 +++++++++++----------------- synapse/util/retryutils.py | 2 +- tests/federation/test_federation_client.py | 12 ++- 7 files changed, 191 insertions(+), 126 deletions(-) create mode 100644 changelog.d/14096.misc (limited to 'synapse/util') diff --git a/changelog.d/14096.misc b/changelog.d/14096.misc new file mode 100644 index 0000000000..2c07dc673b --- /dev/null +++ b/changelog.d/14096.misc @@ -0,0 +1 @@ +Refactor [MSC3030](https://github.com/matrix-org/matrix-spec-proposals/pull/3030) `/timestamp_to_event` endpoint to loop over federation destinations with standard pattern and error handling. diff --git a/synapse/federation/federation_client.py b/synapse/federation/federation_client.py index b220ab43fc..fa225182be 100644 --- a/synapse/federation/federation_client.py +++ b/synapse/federation/federation_client.py @@ -80,6 +80,18 @@ PDU_RETRY_TIME_MS = 1 * 60 * 1000 T = TypeVar("T") +@attr.s(frozen=True, slots=True, auto_attribs=True) +class PulledPduInfo: + """ + A result object that stores the PDU and info about it like which homeserver we + pulled it from (`pull_origin`) + """ + + pdu: EventBase + # Which homeserver we pulled the PDU from + pull_origin: str + + class InvalidResponseError(RuntimeError): """Helper for _try_destination_list: indicates that the server returned a response we couldn't parse @@ -114,7 +126,9 @@ class FederationClient(FederationBase): self.hostname = hs.hostname self.signing_key = hs.signing_key - self._get_pdu_cache: ExpiringCache[str, EventBase] = ExpiringCache( + # Cache mapping `event_id` to a tuple of the event itself and the `pull_origin` + # (which server we pulled the event from) + self._get_pdu_cache: ExpiringCache[str, Tuple[EventBase, str]] = ExpiringCache( cache_name="get_pdu_cache", clock=self._clock, max_len=1000, @@ -352,11 +366,11 @@ class FederationClient(FederationBase): @tag_args async def get_pdu( self, - destinations: Iterable[str], + destinations: Collection[str], event_id: str, room_version: RoomVersion, timeout: Optional[int] = None, - ) -> Optional[EventBase]: + ) -> Optional[PulledPduInfo]: """Requests the PDU with given origin and ID from the remote home servers. @@ -371,11 +385,11 @@ class FederationClient(FederationBase): moving to the next destination. None indicates no timeout. Returns: - The requested PDU, or None if we were unable to find it. + The requested PDU wrapped in `PulledPduInfo`, or None if we were unable to find it. """ logger.debug( - "get_pdu: event_id=%s from destinations=%s", event_id, destinations + "get_pdu(event_id=%s): from destinations=%s", event_id, destinations ) # TODO: Rate limit the number of times we try and get the same event. @@ -384,19 +398,25 @@ class FederationClient(FederationBase): # it gets persisted to the database), so we cache the results of the lookup. # Note that this is separate to the regular get_event cache which caches # events once they have been persisted. - event = self._get_pdu_cache.get(event_id) + get_pdu_cache_entry = self._get_pdu_cache.get(event_id) + event = None + pull_origin = None + if get_pdu_cache_entry: + event, pull_origin = get_pdu_cache_entry # If we don't see the event in the cache, go try to fetch it from the # provided remote federated destinations - if not event: + else: pdu_attempts = self.pdu_destination_tried.setdefault(event_id, {}) + # TODO: We can probably refactor this to use `_try_destination_list` for destination in destinations: now = self._clock.time_msec() last_attempt = pdu_attempts.get(destination, 0) if last_attempt + PDU_RETRY_TIME_MS > now: logger.debug( - "get_pdu: skipping destination=%s because we tried it recently last_attempt=%s and we only check every %s (now=%s)", + "get_pdu(event_id=%s): skipping destination=%s because we tried it recently last_attempt=%s and we only check every %s (now=%s)", + event_id, destination, last_attempt, PDU_RETRY_TIME_MS, @@ -411,43 +431,48 @@ class FederationClient(FederationBase): room_version=room_version, timeout=timeout, ) + pull_origin = destination pdu_attempts[destination] = now if event: # Prime the cache - self._get_pdu_cache[event.event_id] = event + self._get_pdu_cache[event.event_id] = (event, pull_origin) # Now that we have an event, we can break out of this # loop and stop asking other destinations. break + except NotRetryingDestination as e: + logger.info("get_pdu(event_id=%s): %s", event_id, e) + continue + except FederationDeniedError: + logger.info( + "get_pdu(event_id=%s): Not attempting to fetch PDU from %s because the homeserver is not on our federation whitelist", + event_id, + destination, + ) + continue except SynapseError as e: logger.info( - "Failed to get PDU %s from %s because %s", + "get_pdu(event_id=%s): Failed to get PDU from %s because %s", event_id, destination, e, ) continue - except NotRetryingDestination as e: - logger.info(str(e)) - continue - except FederationDeniedError as e: - logger.info(str(e)) - continue except Exception as e: pdu_attempts[destination] = now logger.info( - "Failed to get PDU %s from %s because %s", + "get_pdu(event_id=): Failed to get PDU from %s because %s", event_id, destination, e, ) continue - if not event: + if not event or not pull_origin: return None # `event` now refers to an object stored in `get_pdu_cache`. Our @@ -459,7 +484,7 @@ class FederationClient(FederationBase): event.room_version, ) - return event_copy + return PulledPduInfo(event_copy, pull_origin) @trace @tag_args @@ -699,12 +724,14 @@ class FederationClient(FederationBase): pdu_origin = get_domain_from_id(pdu.sender) if not res and pdu_origin != origin: try: - res = await self.get_pdu( + pulled_pdu_info = await self.get_pdu( destinations=[pdu_origin], event_id=pdu.event_id, room_version=room_version, timeout=10000, ) + if pulled_pdu_info is not None: + res = pulled_pdu_info.pdu except SynapseError: pass @@ -806,6 +833,7 @@ class FederationClient(FederationBase): ) for destination in destinations: + # We don't want to ask our own server for information we don't have if destination == self.server_name: continue @@ -814,9 +842,21 @@ class FederationClient(FederationBase): except ( RequestSendFailed, InvalidResponseError, - NotRetryingDestination, ) as e: logger.warning("Failed to %s via %s: %s", description, destination, e) + # Skip to the next homeserver in the list to try. + continue + except NotRetryingDestination as e: + logger.info("%s: %s", description, e) + continue + except FederationDeniedError: + logger.info( + "%s: Not attempting to %s from %s because the homeserver is not on our federation whitelist", + description, + description, + destination, + ) + continue except UnsupportedRoomVersionError: raise except HttpResponseException as e: @@ -1609,6 +1649,54 @@ class FederationClient(FederationBase): return result async def timestamp_to_event( + self, *, destinations: List[str], room_id: str, timestamp: int, direction: str + ) -> Optional["TimestampToEventResponse"]: + """ + Calls each remote federating server from `destinations` asking for their closest + event to the given timestamp in the given direction until we get a response. + Also validates the response to always return the expected keys or raises an + error. + + Args: + destinations: The domains of homeservers to try fetching from + room_id: Room to fetch the event from + timestamp: The point in time (inclusive) we should navigate from in + the given direction to find the closest event. + direction: ["f"|"b"] to indicate whether we should navigate forward + or backward from the given timestamp to find the closest event. + + Returns: + A parsed TimestampToEventResponse including the closest event_id + and origin_server_ts or None if no destination has a response. + """ + + async def _timestamp_to_event_from_destination( + destination: str, + ) -> TimestampToEventResponse: + return await self._timestamp_to_event_from_destination( + destination, room_id, timestamp, direction + ) + + try: + # Loop through each homeserver candidate until we get a succesful response + timestamp_to_event_response = await self._try_destination_list( + "timestamp_to_event", + destinations, + # TODO: The requested timestamp may lie in a part of the + # event graph that the remote server *also* didn't have, + # in which case they will have returned another event + # which may be nowhere near the requested timestamp. In + # the future, we may need to reconcile that gap and ask + # other homeservers, and/or extend `/timestamp_to_event` + # to return events on *both* sides of the timestamp to + # help reconcile the gap faster. + _timestamp_to_event_from_destination, + ) + return timestamp_to_event_response + except SynapseError: + return None + + async def _timestamp_to_event_from_destination( self, destination: str, room_id: str, timestamp: int, direction: str ) -> "TimestampToEventResponse": """ diff --git a/synapse/handlers/federation.py b/synapse/handlers/federation.py index 4fbc79a6cb..5fc3b8bc8c 100644 --- a/synapse/handlers/federation.py +++ b/synapse/handlers/federation.py @@ -442,6 +442,15 @@ class FederationHandler: # appropriate stuff. # TODO: We can probably do something more intelligent here. return True + except NotRetryingDestination as e: + logger.info("_maybe_backfill_inner: %s", e) + continue + except FederationDeniedError: + logger.info( + "_maybe_backfill_inner: Not attempting to backfill from %s because the homeserver is not on our federation whitelist", + dom, + ) + continue except (SynapseError, InvalidResponseError) as e: logger.info("Failed to backfill from %s because %s", dom, e) continue @@ -477,15 +486,9 @@ class FederationHandler: logger.info("Failed to backfill from %s because %s", dom, e) continue - except NotRetryingDestination as e: - logger.info(str(e)) - continue except RequestSendFailed as e: logger.info("Failed to get backfill from %s because %s", dom, e) continue - except FederationDeniedError as e: - logger.info(e) - continue except Exception as e: logger.exception("Failed to backfill from %s because %s", dom, e) continue diff --git a/synapse/handlers/federation_event.py b/synapse/handlers/federation_event.py index 7da6316a82..9ca5df7c78 100644 --- a/synapse/handlers/federation_event.py +++ b/synapse/handlers/federation_event.py @@ -58,7 +58,7 @@ from synapse.event_auth import ( ) from synapse.events import EventBase from synapse.events.snapshot import EventContext -from synapse.federation.federation_client import InvalidResponseError +from synapse.federation.federation_client import InvalidResponseError, PulledPduInfo from synapse.logging.context import nested_logging_context from synapse.logging.opentracing import ( SynapseTags, @@ -1517,8 +1517,8 @@ class FederationEventHandler: ) async def backfill_event_id( - self, destination: str, room_id: str, event_id: str - ) -> EventBase: + self, destinations: List[str], room_id: str, event_id: str + ) -> PulledPduInfo: """Backfill a single event and persist it as a non-outlier which means we also pull in all of the state and auth events necessary for it. @@ -1530,24 +1530,21 @@ class FederationEventHandler: Raises: FederationError if we are unable to find the event from the destination """ - logger.info( - "backfill_event_id: event_id=%s from destination=%s", event_id, destination - ) + logger.info("backfill_event_id: event_id=%s", event_id) room_version = await self._store.get_room_version(room_id) - event_from_response = await self._federation_client.get_pdu( - [destination], + pulled_pdu_info = await self._federation_client.get_pdu( + destinations, event_id, room_version, ) - if not event_from_response: + if not pulled_pdu_info: raise FederationError( "ERROR", 404, - "Unable to find event_id=%s from destination=%s to backfill." - % (event_id, destination), + f"Unable to find event_id={event_id} from remote servers to backfill.", affected=event_id, ) @@ -1555,13 +1552,13 @@ class FederationEventHandler: # and auth events to de-outlier it. This also sets up the necessary # `state_groups` for the event. await self._process_pulled_events( - destination, - [event_from_response], + pulled_pdu_info.pull_origin, + [pulled_pdu_info.pdu], # Prevent notifications going to clients backfilled=True, ) - return event_from_response + return pulled_pdu_info @trace @tag_args @@ -1584,19 +1581,19 @@ class FederationEventHandler: async def get_event(event_id: str) -> None: with nested_logging_context(event_id): try: - event = await self._federation_client.get_pdu( + pulled_pdu_info = await self._federation_client.get_pdu( [destination], event_id, room_version, ) - if event is None: + if pulled_pdu_info is None: logger.warning( "Server %s didn't return event %s", destination, event_id, ) return - events.append(event) + events.append(pulled_pdu_info.pdu) except Exception as e: logger.warning( diff --git a/synapse/handlers/room.py b/synapse/handlers/room.py index cc1e5c8f97..de97886ea9 100644 --- a/synapse/handlers/room.py +++ b/synapse/handlers/room.py @@ -49,7 +49,6 @@ from synapse.api.constants import ( from synapse.api.errors import ( AuthError, Codes, - HttpResponseException, LimitExceededError, NotFoundError, StoreError, @@ -60,7 +59,6 @@ from synapse.api.room_versions import KNOWN_ROOM_VERSIONS, RoomVersion from synapse.event_auth import validate_event_for_room_version from synapse.events import EventBase from synapse.events.utils import copy_and_fixup_power_levels_contents -from synapse.federation.federation_client import InvalidResponseError from synapse.handlers.relations import BundledAggregations from synapse.module_api import NOT_SPAM from synapse.rest.admin._base import assert_user_is_admin @@ -1472,7 +1470,12 @@ class TimestampLookupHandler: Raises: SynapseError if unable to find any event locally in the given direction """ - + logger.debug( + "get_event_for_timestamp(room_id=%s, timestamp=%s, direction=%s) Finding closest event...", + room_id, + timestamp, + direction, + ) local_event_id = await self.store.get_event_id_for_timestamp( room_id, timestamp, direction ) @@ -1524,85 +1527,54 @@ class TimestampLookupHandler: ) ) - # Loop through each homeserver candidate until we get a succesful response - for domain in likely_domains: - # We don't want to ask our own server for information we don't have - if domain == self.server_name: - continue + remote_response = await self.federation_client.timestamp_to_event( + destinations=likely_domains, + room_id=room_id, + timestamp=timestamp, + direction=direction, + ) + if remote_response is not None: + logger.debug( + "get_event_for_timestamp: remote_response=%s", + remote_response, + ) - try: - remote_response = await self.federation_client.timestamp_to_event( - domain, room_id, timestamp, direction - ) - logger.debug( - "get_event_for_timestamp: response from domain(%s)=%s", - domain, - remote_response, - ) + remote_event_id = remote_response.event_id + remote_origin_server_ts = remote_response.origin_server_ts - remote_event_id = remote_response.event_id - remote_origin_server_ts = remote_response.origin_server_ts - - # Backfill this event so we can get a pagination token for - # it with `/context` and paginate `/messages` from this - # point. - # - # TODO: The requested timestamp may lie in a part of the - # event graph that the remote server *also* didn't have, - # in which case they will have returned another event - # which may be nowhere near the requested timestamp. In - # the future, we may need to reconcile that gap and ask - # other homeservers, and/or extend `/timestamp_to_event` - # to return events on *both* sides of the timestamp to - # help reconcile the gap faster. - remote_event = ( - await self.federation_event_handler.backfill_event_id( - domain, room_id, remote_event_id - ) - ) + # Backfill this event so we can get a pagination token for + # it with `/context` and paginate `/messages` from this + # point. + pulled_pdu_info = await self.federation_event_handler.backfill_event_id( + likely_domains, room_id, remote_event_id + ) + remote_event = pulled_pdu_info.pdu - # XXX: When we see that the remote server is not trustworthy, - # maybe we should not ask them first in the future. - if remote_origin_server_ts != remote_event.origin_server_ts: - logger.info( - "get_event_for_timestamp: Remote server (%s) claimed that remote_event_id=%s occured at remote_origin_server_ts=%s but that isn't true (actually occured at %s). Their claims are dubious and we should consider not trusting them.", - domain, - remote_event_id, - remote_origin_server_ts, - remote_event.origin_server_ts, - ) - - # Only return the remote event if it's closer than the local event - if not local_event or ( - abs(remote_event.origin_server_ts - timestamp) - < abs(local_event.origin_server_ts - timestamp) - ): - logger.info( - "get_event_for_timestamp: returning remote_event_id=%s (%s) since it's closer to timestamp=%s than local_event=%s (%s)", - remote_event_id, - remote_event.origin_server_ts, - timestamp, - local_event.event_id if local_event else None, - local_event.origin_server_ts if local_event else None, - ) - return remote_event_id, remote_origin_server_ts - except (HttpResponseException, InvalidResponseError) as ex: - # Let's not put a high priority on some other homeserver - # failing to respond or giving a random response - logger.debug( - "get_event_for_timestamp: Failed to fetch /timestamp_to_event from %s because of exception(%s) %s args=%s", - domain, - type(ex).__name__, - ex, - ex.args, + # XXX: When we see that the remote server is not trustworthy, + # maybe we should not ask them first in the future. + if remote_origin_server_ts != remote_event.origin_server_ts: + logger.info( + "get_event_for_timestamp: Remote server (%s) claimed that remote_event_id=%s occured at remote_origin_server_ts=%s but that isn't true (actually occured at %s). Their claims are dubious and we should consider not trusting them.", + pulled_pdu_info.pull_origin, + remote_event_id, + remote_origin_server_ts, + remote_event.origin_server_ts, ) - except Exception: - # But we do want to see some exceptions in our code - logger.warning( - "get_event_for_timestamp: Failed to fetch /timestamp_to_event from %s because of exception", - domain, - exc_info=True, + + # Only return the remote event if it's closer than the local event + if not local_event or ( + abs(remote_event.origin_server_ts - timestamp) + < abs(local_event.origin_server_ts - timestamp) + ): + logger.info( + "get_event_for_timestamp: returning remote_event_id=%s (%s) since it's closer to timestamp=%s than local_event=%s (%s)", + remote_event_id, + remote_event.origin_server_ts, + timestamp, + local_event.event_id if local_event else None, + local_event.origin_server_ts if local_event else None, ) + return remote_event_id, remote_origin_server_ts # To appease mypy, we have to add both of these conditions to check for # `None`. We only expect `local_event` to be `None` when diff --git a/synapse/util/retryutils.py b/synapse/util/retryutils.py index d0a69ff843..dcc037b982 100644 --- a/synapse/util/retryutils.py +++ b/synapse/util/retryutils.py @@ -51,7 +51,7 @@ class NotRetryingDestination(Exception): destination: the domain in question """ - msg = "Not retrying server %s." % (destination,) + msg = f"Not retrying server {destination} because we tried it recently retry_last_ts={retry_last_ts} and we won't check for another retry_interval={retry_interval}ms." super().__init__(msg) self.retry_last_ts = retry_last_ts diff --git a/tests/federation/test_federation_client.py b/tests/federation/test_federation_client.py index 51d3bb8fff..e67f405826 100644 --- a/tests/federation/test_federation_client.py +++ b/tests/federation/test_federation_client.py @@ -142,14 +142,14 @@ class FederationClientTest(FederatingHomeserverTestCase): def test_get_pdu_returns_nothing_when_event_does_not_exist(self): """No event should be returned when the event does not exist""" - remote_pdu = self.get_success( + pulled_pdu_info = self.get_success( self.hs.get_federation_client().get_pdu( ["yet.another.server"], "event_should_not_exist", RoomVersions.V9, ) ) - self.assertEqual(remote_pdu, None) + self.assertEqual(pulled_pdu_info, None) def test_get_pdu(self): """Test to make sure an event is returned by `get_pdu()`""" @@ -169,13 +169,15 @@ class FederationClientTest(FederatingHomeserverTestCase): remote_pdu.internal_metadata.outlier = True # Get the event again. This time it should read it from cache. - remote_pdu2 = self.get_success( + pulled_pdu_info2 = self.get_success( self.hs.get_federation_client().get_pdu( ["yet.another.server"], remote_pdu.event_id, RoomVersions.V9, ) ) + self.assertIsNotNone(pulled_pdu_info2) + remote_pdu2 = pulled_pdu_info2.pdu # Sanity check that we are working against the same event self.assertEqual(remote_pdu.event_id, remote_pdu2.event_id) @@ -215,13 +217,15 @@ class FederationClientTest(FederatingHomeserverTestCase): ) ) - remote_pdu = self.get_success( + pulled_pdu_info = self.get_success( self.hs.get_federation_client().get_pdu( ["yet.another.server"], "event_id", RoomVersions.V9, ) ) + self.assertIsNotNone(pulled_pdu_info) + remote_pdu = pulled_pdu_info.pdu # check the right call got made to the agent self._mock_agent.request.assert_called_once_with( -- cgit 1.5.1 From 13ca8bb2fc05d338ccf62e6f8d1cbf5021d935ba Mon Sep 17 00:00:00 2001 From: Patrick Cloke Date: Thu, 10 Nov 2022 15:33:34 -0500 Subject: Remove duplicated code to evict entries. (#14410) This code was factored out to a method, but also left in-place. Calling this twice in a row makes no sense: the first call will reduce the size appropriately, but the loop will immediately exit since the cache size was already reduced. --- changelog.d/14410.misc | 1 + synapse/util/caches/stream_change_cache.py | 11 ++--------- 2 files changed, 3 insertions(+), 9 deletions(-) create mode 100644 changelog.d/14410.misc (limited to 'synapse/util') diff --git a/changelog.d/14410.misc b/changelog.d/14410.misc new file mode 100644 index 0000000000..f085a8bfb2 --- /dev/null +++ b/changelog.d/14410.misc @@ -0,0 +1 @@ +Remove unreachable code. diff --git a/synapse/util/caches/stream_change_cache.py b/synapse/util/caches/stream_change_cache.py index 330709b8b7..666f4b6895 100644 --- a/synapse/util/caches/stream_change_cache.py +++ b/synapse/util/caches/stream_change_cache.py @@ -72,7 +72,7 @@ class StreamChangeCache: items from the cache. Returns: - bool: Whether the cache changed size or not. + Whether the cache changed size or not. """ new_size = math.floor(self._original_max_size * factor) if new_size != self._max_size: @@ -188,14 +188,8 @@ class StreamChangeCache: self._entity_to_key[entity] = stream_pos self._evict() - # if the cache is too big, remove entries - while len(self._cache) > self._max_size: - k, r = self._cache.popitem(0) - self._earliest_known_stream_pos = max(k, self._earliest_known_stream_pos) - for entity in r: - del self._entity_to_key[entity] - def _evict(self) -> None: + # if the cache is too big, remove entries while len(self._cache) > self._max_size: k, r = self._cache.popitem(0) self._earliest_known_stream_pos = max(k, self._earliest_known_stream_pos) @@ -203,7 +197,6 @@ class StreamChangeCache: self._entity_to_key.pop(entity, None) def get_max_pos_of_last_change(self, entity: EntityType) -> int: - """Returns an upper bound of the stream id of the last change to an entity. """ -- cgit 1.5.1 From d8cc86eff484b6f570f55a5badb337080c6e4dcd Mon Sep 17 00:00:00 2001 From: Patrick Cloke Date: Wed, 16 Nov 2022 10:25:24 -0500 Subject: Remove redundant types from comments. (#14412) Remove type hints from comments which have been added as Python type hints. This helps avoid drift between comments and reality, as well as removing redundant information. Also adds some missing type hints which were simple to fill in. --- changelog.d/14412.misc | 1 + synapse/api/errors.py | 2 +- synapse/config/logger.py | 5 ++- synapse/crypto/keyring.py | 9 +++-- synapse/events/__init__.py | 3 +- synapse/federation/transport/client.py | 11 +++--- synapse/federation/transport/server/_base.py | 4 +-- synapse/handlers/e2e_keys.py | 2 +- synapse/handlers/e2e_room_keys.py | 5 +-- synapse/handlers/federation.py | 4 +-- synapse/handlers/identity.py | 2 +- synapse/handlers/oidc.py | 2 +- synapse/handlers/presence.py | 2 +- synapse/handlers/saml.py | 4 +-- synapse/http/additional_resource.py | 3 +- synapse/http/federation/matrix_federation_agent.py | 9 +++-- synapse/http/matrixfederationclient.py | 3 +- synapse/http/proxyagent.py | 20 +++++------ synapse/http/server.py | 2 +- synapse/http/site.py | 2 +- synapse/logging/context.py | 39 +++++++++++----------- synapse/logging/opentracing.py | 4 +-- synapse/module_api/__init__.py | 7 ++-- synapse/replication/http/_base.py | 2 +- synapse/rest/admin/users.py | 5 +-- synapse/rest/client/login.py | 2 +- synapse/rest/media/v1/media_repository.py | 4 +-- synapse/rest/media/v1/thumbnailer.py | 4 +-- synapse/server_notices/consent_server_notices.py | 5 ++- .../resource_limits_server_notices.py | 12 ++++--- synapse/storage/controllers/persist_events.py | 5 ++- synapse/storage/databases/main/devices.py | 2 +- synapse/storage/databases/main/e2e_room_keys.py | 8 ++--- synapse/storage/databases/main/end_to_end_keys.py | 7 ++-- synapse/storage/databases/main/events.py | 22 ++++++------ synapse/storage/databases/main/events_worker.py | 2 +- .../storage/databases/main/monthly_active_users.py | 8 ++--- synapse/storage/databases/main/registration.py | 6 ++-- synapse/storage/databases/main/room.py | 8 +++-- synapse/storage/databases/main/user_directory.py | 9 +++-- synapse/types.py | 4 +-- synapse/util/async_helpers.py | 3 +- synapse/util/caches/__init__.py | 2 +- synapse/util/caches/deferred_cache.py | 2 +- synapse/util/caches/dictionary_cache.py | 9 ++--- synapse/util/caches/expiringcache.py | 2 +- synapse/util/caches/lrucache.py | 8 ++--- synapse/util/ratelimitutils.py | 2 +- synapse/util/threepids.py | 2 +- synapse/util/wheel_timer.py | 4 +-- tests/http/__init__.py | 7 ++-- tests/replication/slave/storage/test_events.py | 7 ++-- tests/replication/test_multi_media_repo.py | 14 ++++---- .../test_resource_limits_server_notices.py | 10 +++--- tests/unittest.py | 18 +++++----- 55 files changed, 174 insertions(+), 176 deletions(-) create mode 100644 changelog.d/14412.misc (limited to 'synapse/util') diff --git a/changelog.d/14412.misc b/changelog.d/14412.misc new file mode 100644 index 0000000000..4da061d461 --- /dev/null +++ b/changelog.d/14412.misc @@ -0,0 +1 @@ +Remove duplicated type information from type hints. diff --git a/synapse/api/errors.py b/synapse/api/errors.py index 400dd12aba..e2cfcea0f2 100644 --- a/synapse/api/errors.py +++ b/synapse/api/errors.py @@ -713,7 +713,7 @@ class HttpResponseException(CodeMessageException): set to the reason code from the HTTP response. Returns: - SynapseError: + The error converted to a SynapseError. """ # try to parse the body as json, to get better errcode/msg, but # default to M_UNKNOWN with the HTTP status as the error text diff --git a/synapse/config/logger.py b/synapse/config/logger.py index 94d1150415..5468b963a2 100644 --- a/synapse/config/logger.py +++ b/synapse/config/logger.py @@ -317,10 +317,9 @@ def setup_logging( Set up the logging subsystem. Args: - config (LoggingConfig | synapse.config.worker.WorkerConfig): - configuration data + config: configuration data - use_worker_options (bool): True to use the 'worker_log_config' option + use_worker_options: True to use the 'worker_log_config' option instead of 'log_config'. logBeginner: The Twisted logBeginner to use. diff --git a/synapse/crypto/keyring.py b/synapse/crypto/keyring.py index c88afb2986..dd9b8089ec 100644 --- a/synapse/crypto/keyring.py +++ b/synapse/crypto/keyring.py @@ -213,7 +213,7 @@ class Keyring: def verify_json_objects_for_server( self, server_and_json: Iterable[Tuple[str, dict, int]] - ) -> List[defer.Deferred]: + ) -> List["defer.Deferred[None]"]: """Bulk verifies signatures of json objects, bulk fetching keys as necessary. @@ -226,10 +226,9 @@ class Keyring: valid. Returns: - List: for each input triplet, a deferred indicating success - or failure to verify each json object's signature for the given - server_name. The deferreds run their callbacks in the sentinel - logcontext. + For each input triplet, a deferred indicating success or failure to + verify each json object's signature for the given server_name. The + deferreds run their callbacks in the sentinel logcontext. """ return [ run_in_background( diff --git a/synapse/events/__init__.py b/synapse/events/__init__.py index 030c3ca408..8aca9a3ab9 100644 --- a/synapse/events/__init__.py +++ b/synapse/events/__init__.py @@ -597,8 +597,7 @@ def _event_type_from_format_version( format_version: The event format version Returns: - type: A type that can be initialized as per the initializer of - `FrozenEvent` + A type that can be initialized as per the initializer of `FrozenEvent` """ if format_version == EventFormatVersions.ROOM_V1_V2: diff --git a/synapse/federation/transport/client.py b/synapse/federation/transport/client.py index cd39d4d111..a3cfc701cd 100644 --- a/synapse/federation/transport/client.py +++ b/synapse/federation/transport/client.py @@ -280,12 +280,11 @@ class TransportLayerClient: Note that this does not append any events to any graphs. Args: - destination (str): address of remote homeserver - room_id (str): room to join/leave - user_id (str): user to be joined/left - membership (str): one of join/leave - params (dict[str, str|Iterable[str]]): Query parameters to include in the - request. + destination: address of remote homeserver + room_id: room to join/leave + user_id: user to be joined/left + membership: one of join/leave + params: Query parameters to include in the request. Returns: Succeeds when we get a 2xx HTTP response. The result diff --git a/synapse/federation/transport/server/_base.py b/synapse/federation/transport/server/_base.py index 1db8009d6c..cdaf0d5de7 100644 --- a/synapse/federation/transport/server/_base.py +++ b/synapse/federation/transport/server/_base.py @@ -224,10 +224,10 @@ class BaseFederationServlet: With arguments: - origin (unicode|None): The authenticated server_name of the calling server, + origin (str|None): The authenticated server_name of the calling server, unless REQUIRE_AUTH is set to False and authentication failed. - content (unicode|None): decoded json body of the request. None if the + content (str|None): decoded json body of the request. None if the request was a GET. query (dict[bytes, list[bytes]]): Query params from the request. url-decoded diff --git a/synapse/handlers/e2e_keys.py b/synapse/handlers/e2e_keys.py index a9912c467d..bf1221f523 100644 --- a/synapse/handlers/e2e_keys.py +++ b/synapse/handlers/e2e_keys.py @@ -870,7 +870,7 @@ class E2eKeysHandler: - signatures of the user's master key by the user's devices. Args: - user_id (string): the user uploading the keys + user_id: the user uploading the keys signatures (dict[string, dict]): map of devices to signed keys Returns: diff --git a/synapse/handlers/e2e_room_keys.py b/synapse/handlers/e2e_room_keys.py index 28dc08c22a..83f53ceb88 100644 --- a/synapse/handlers/e2e_room_keys.py +++ b/synapse/handlers/e2e_room_keys.py @@ -377,8 +377,9 @@ class E2eRoomKeysHandler: """Deletes a given version of the user's e2e_room_keys backup Args: - user_id(str): the user whose current backup version we're deleting - version(str): the version id of the backup being deleted + user_id: the user whose current backup version we're deleting + version: Optional. the version ID of the backup version we're deleting + If missing, we delete the current backup version info. Raises: NotFoundError: if this backup version doesn't exist """ diff --git a/synapse/handlers/federation.py b/synapse/handlers/federation.py index 5fc3b8bc8c..188f0956ef 100644 --- a/synapse/handlers/federation.py +++ b/synapse/handlers/federation.py @@ -1596,8 +1596,8 @@ class FederationHandler: Fetch the complexity of a remote room over federation. Args: - remote_room_hosts (list[str]): The remote servers to ask. - room_id (str): The room ID to ask about. + remote_room_hosts: The remote servers to ask. + room_id: The room ID to ask about. Returns: Dict contains the complexity diff --git a/synapse/handlers/identity.py b/synapse/handlers/identity.py index 93d09e9939..848e46eb9b 100644 --- a/synapse/handlers/identity.py +++ b/synapse/handlers/identity.py @@ -711,7 +711,7 @@ class IdentityHandler: inviter_display_name: The current display name of the inviter. inviter_avatar_url: The URL of the inviter's avatar. - id_access_token (str): The access token to authenticate to the identity + id_access_token: The access token to authenticate to the identity server with Returns: diff --git a/synapse/handlers/oidc.py b/synapse/handlers/oidc.py index 867973dcca..41c675f408 100644 --- a/synapse/handlers/oidc.py +++ b/synapse/handlers/oidc.py @@ -787,7 +787,7 @@ class OidcProvider: Must include an ``access_token`` field. Returns: - UserInfo: an object representing the user. + an object representing the user. """ logger.debug("Using the OAuth2 access_token to request userinfo") metadata = await self.load_metadata() diff --git a/synapse/handlers/presence.py b/synapse/handlers/presence.py index 0066d63987..b7bc787636 100644 --- a/synapse/handlers/presence.py +++ b/synapse/handlers/presence.py @@ -201,7 +201,7 @@ class BasePresenceHandler(abc.ABC): """Get the current presence state for multiple users. Returns: - dict: `user_id` -> `UserPresenceState` + A mapping of `user_id` -> `UserPresenceState` """ states = {} missing = [] diff --git a/synapse/handlers/saml.py b/synapse/handlers/saml.py index 9602f0d0bb..874860d461 100644 --- a/synapse/handlers/saml.py +++ b/synapse/handlers/saml.py @@ -441,7 +441,7 @@ class DefaultSamlMappingProvider: client_redirect_url: where the client wants to redirect to Returns: - dict: A dict containing new user attributes. Possible keys: + A dict containing new user attributes. Possible keys: * mxid_localpart (str): Required. The localpart of the user's mxid * displayname (str): The displayname of the user * emails (list[str]): Any emails for the user @@ -483,7 +483,7 @@ class DefaultSamlMappingProvider: Args: config: A dictionary containing configuration options for this provider Returns: - SamlConfig: A custom config object for this module + A custom config object for this module """ # Parse config options and use defaults where necessary mxid_source_attribute = config.get("mxid_source_attribute", "uid") diff --git a/synapse/http/additional_resource.py b/synapse/http/additional_resource.py index 6a9f6635d2..8729630581 100644 --- a/synapse/http/additional_resource.py +++ b/synapse/http/additional_resource.py @@ -45,8 +45,7 @@ class AdditionalResource(DirectServeJsonResource): Args: hs: homeserver - handler ((twisted.web.server.Request) -> twisted.internet.defer.Deferred): - function to be called to handle the request. + handler: function to be called to handle the request. """ super().__init__() self._handler = handler diff --git a/synapse/http/federation/matrix_federation_agent.py b/synapse/http/federation/matrix_federation_agent.py index 2f0177f1e2..0359231e7d 100644 --- a/synapse/http/federation/matrix_federation_agent.py +++ b/synapse/http/federation/matrix_federation_agent.py @@ -155,11 +155,10 @@ class MatrixFederationAgent: a file for a file upload). Or None if the request is to have no body. Returns: - Deferred[twisted.web.iweb.IResponse]: - fires when the header of the response has been received (regardless of the - response status code). Fails if there is any problem which prevents that - response from being received (including problems that prevent the request - from being sent). + A deferred which fires when the header of the response has been received + (regardless of the response status code). Fails if there is any problem + which prevents that response from being received (including problems that + prevent the request from being sent). """ # We use urlparse as that will set `port` to None if there is no # explicit port. diff --git a/synapse/http/matrixfederationclient.py b/synapse/http/matrixfederationclient.py index 3c35b1d2c7..b92f1d3d1a 100644 --- a/synapse/http/matrixfederationclient.py +++ b/synapse/http/matrixfederationclient.py @@ -951,8 +951,7 @@ class MatrixFederationHttpClient: args: query params Returns: - dict|list: Succeeds when we get a 2xx HTTP response. The - result will be the decoded JSON body. + Succeeds when we get a 2xx HTTP response. The result will be the decoded JSON body. Raises: HttpResponseException: If we get an HTTP response code >= 300 diff --git a/synapse/http/proxyagent.py b/synapse/http/proxyagent.py index 1f8227896f..18899bc6d1 100644 --- a/synapse/http/proxyagent.py +++ b/synapse/http/proxyagent.py @@ -34,7 +34,7 @@ from twisted.web.client import ( ) from twisted.web.error import SchemeNotSupported from twisted.web.http_headers import Headers -from twisted.web.iweb import IAgent, IBodyProducer, IPolicyForHTTPS +from twisted.web.iweb import IAgent, IBodyProducer, IPolicyForHTTPS, IResponse from synapse.http import redact_uri from synapse.http.connectproxyclient import HTTPConnectProxyEndpoint, ProxyCredentials @@ -134,7 +134,7 @@ class ProxyAgent(_AgentBase): uri: bytes, headers: Optional[Headers] = None, bodyProducer: Optional[IBodyProducer] = None, - ) -> defer.Deferred: + ) -> "defer.Deferred[IResponse]": """ Issue a request to the server indicated by the given uri. @@ -157,17 +157,17 @@ class ProxyAgent(_AgentBase): a file upload). Or, None if the request is to have no body. Returns: - Deferred[IResponse]: completes when the header of the response has - been received (regardless of the response status code). + A deferred which completes when the header of the response has + been received (regardless of the response status code). - Can fail with: - SchemeNotSupported: if the uri is not http or https + Can fail with: + SchemeNotSupported: if the uri is not http or https - twisted.internet.error.TimeoutError if the server we are connecting - to (proxy or destination) does not accept a connection before - connectTimeout. + twisted.internet.error.TimeoutError if the server we are connecting + to (proxy or destination) does not accept a connection before + connectTimeout. - ... other things too. + ... other things too. """ uri = uri.strip() if not _VALID_URI.match(uri): diff --git a/synapse/http/server.py b/synapse/http/server.py index b26e34bceb..051a1899a0 100644 --- a/synapse/http/server.py +++ b/synapse/http/server.py @@ -267,7 +267,7 @@ class HttpServer(Protocol): request. The first argument will be the request object and subsequent arguments will be any matched groups from the regex. This should return either tuple of (code, response), or None. - servlet_classname (str): The name of the handler to be used in prometheus + servlet_classname: The name of the handler to be used in prometheus and opentracing logs. """ diff --git a/synapse/http/site.py b/synapse/http/site.py index 3dbd541fed..6a1dbf7f33 100644 --- a/synapse/http/site.py +++ b/synapse/http/site.py @@ -400,7 +400,7 @@ class SynapseRequest(Request): be sure to call finished_processing. Args: - servlet_name (str): the name of the servlet which will be + servlet_name: the name of the servlet which will be processing this request. This is used in the metrics. It is possible to update this afterwards by updating diff --git a/synapse/logging/context.py b/synapse/logging/context.py index 6a08ffed64..f62bea968f 100644 --- a/synapse/logging/context.py +++ b/synapse/logging/context.py @@ -117,8 +117,7 @@ class ContextResourceUsage: """Create a new ContextResourceUsage Args: - copy_from (ContextResourceUsage|None): if not None, an object to - copy stats from + copy_from: if not None, an object to copy stats from """ if copy_from is None: self.reset() @@ -162,7 +161,7 @@ class ContextResourceUsage: """Add another ContextResourceUsage's stats to this one's. Args: - other (ContextResourceUsage): the other resource usage object + other: the other resource usage object """ self.ru_utime += other.ru_utime self.ru_stime += other.ru_stime @@ -342,7 +341,7 @@ class LoggingContext: called directly. Returns: - LoggingContext: the current logging context + The current logging context """ warnings.warn( "synapse.logging.context.LoggingContext.current_context() is deprecated " @@ -362,7 +361,8 @@ class LoggingContext: called directly. Args: - context(LoggingContext): The context to activate. + context: The context to activate. + Returns: The context that was previously active """ @@ -474,8 +474,7 @@ class LoggingContext: """Get resources used by this logcontext so far. Returns: - ContextResourceUsage: a *copy* of the object tracking resource - usage so far + A *copy* of the object tracking resource usage so far """ # we always return a copy, for consistency res = self._resource_usage.copy() @@ -663,7 +662,8 @@ def current_context() -> LoggingContextOrSentinel: def set_current_context(context: LoggingContextOrSentinel) -> LoggingContextOrSentinel: """Set the current logging context in thread local storage Args: - context(LoggingContext): The context to activate. + context: The context to activate. + Returns: The context that was previously active """ @@ -700,7 +700,7 @@ def nested_logging_context(suffix: str) -> LoggingContext: suffix: suffix to add to the parent context's 'name'. Returns: - LoggingContext: new logging context. + A new logging context. """ curr_context = current_context() if not curr_context: @@ -898,20 +898,19 @@ def defer_to_thread( on it. Args: - reactor (twisted.internet.base.ReactorBase): The reactor in whose main thread - the Deferred will be invoked, and whose threadpool we should use for the - function. + reactor: The reactor in whose main thread the Deferred will be invoked, + and whose threadpool we should use for the function. Normally this will be hs.get_reactor(). - f (callable): The function to call. + f: The function to call. args: positional arguments to pass to f. kwargs: keyword arguments to pass to f. Returns: - Deferred: A Deferred which fires a callback with the result of `f`, or an + A Deferred which fires a callback with the result of `f`, or an errback if `f` throws an exception. """ return defer_to_threadpool(reactor, reactor.getThreadPool(), f, *args, **kwargs) @@ -939,20 +938,20 @@ def defer_to_threadpool( on it. Args: - reactor (twisted.internet.base.ReactorBase): The reactor in whose main thread - the Deferred will be invoked. Normally this will be hs.get_reactor(). + reactor: The reactor in whose main thread the Deferred will be invoked. + Normally this will be hs.get_reactor(). - threadpool (twisted.python.threadpool.ThreadPool): The threadpool to use for - running `f`. Normally this will be hs.get_reactor().getThreadPool(). + threadpool: The threadpool to use for running `f`. Normally this will be + hs.get_reactor().getThreadPool(). - f (callable): The function to call. + f: The function to call. args: positional arguments to pass to f. kwargs: keyword arguments to pass to f. Returns: - Deferred: A Deferred which fires a callback with the result of `f`, or an + A Deferred which fires a callback with the result of `f`, or an errback if `f` throws an exception. """ curr_context = current_context() diff --git a/synapse/logging/opentracing.py b/synapse/logging/opentracing.py index 8ce5a2a338..b69060854f 100644 --- a/synapse/logging/opentracing.py +++ b/synapse/logging/opentracing.py @@ -721,7 +721,7 @@ def inject_header_dict( destination: address of entity receiving the span context. Must be given unless check_destination is False. The context will only be injected if the destination matches the opentracing whitelist - check_destination (bool): If false, destination will be ignored and the context + check_destination: If false, destination will be ignored and the context will always be injected. Note: @@ -780,7 +780,7 @@ def get_active_span_text_map(destination: Optional[str] = None) -> Dict[str, str destination: the name of the remote server. Returns: - dict: the active span's context if opentracing is enabled, otherwise empty. + the active span's context if opentracing is enabled, otherwise empty. """ if destination and not whitelisted_homeserver(destination): diff --git a/synapse/module_api/__init__.py b/synapse/module_api/__init__.py index 30e689d00d..1adc1fd64f 100644 --- a/synapse/module_api/__init__.py +++ b/synapse/module_api/__init__.py @@ -787,7 +787,7 @@ class ModuleApi: Added in Synapse v0.25.0. Args: - access_token(str): access token + access_token: access token Returns: twisted.internet.defer.Deferred - resolves once the access token @@ -832,7 +832,7 @@ class ModuleApi: **kwargs: named args to be passed to func Returns: - Deferred[object]: result of func + Result of func """ # type-ignore: See https://github.com/python/mypy/issues/8862 return defer.ensureDeferred( @@ -924,8 +924,7 @@ class ModuleApi: to represent 'any') of the room state to acquire. Returns: - twisted.internet.defer.Deferred[list(synapse.events.FrozenEvent)]: - The filtered state events in the room. + The filtered state events in the room. """ state_ids = yield defer.ensureDeferred( self._storage_controllers.state.get_current_state_ids( diff --git a/synapse/replication/http/_base.py b/synapse/replication/http/_base.py index 5e661f8c73..3f4d3fc51a 100644 --- a/synapse/replication/http/_base.py +++ b/synapse/replication/http/_base.py @@ -153,7 +153,7 @@ class ReplicationEndpoint(metaclass=abc.ABCMeta): argument list. Returns: - dict: If POST/PUT request then dictionary must be JSON serialisable, + If POST/PUT request then dictionary must be JSON serialisable, otherwise must be appropriate for adding as query args. """ return {} diff --git a/synapse/rest/admin/users.py b/synapse/rest/admin/users.py index 1951b8a9f2..6e0c44be2a 100644 --- a/synapse/rest/admin/users.py +++ b/synapse/rest/admin/users.py @@ -903,8 +903,9 @@ class PushersRestServlet(RestServlet): @user:server/pushers Returns: - pushers: Dictionary containing pushers information. - total: Number of pushers in dictionary `pushers`. + A dictionary with keys: + pushers: Dictionary containing pushers information. + total: Number of pushers in dictionary `pushers`. """ PATTERNS = admin_patterns("/users/(?P[^/]*)/pushers$") diff --git a/synapse/rest/client/login.py b/synapse/rest/client/login.py index 05706b598c..8adced41e5 100644 --- a/synapse/rest/client/login.py +++ b/synapse/rest/client/login.py @@ -350,7 +350,7 @@ class LoginRestServlet(RestServlet): auth_provider_session_id: The session ID got during login from the SSO IdP. Returns: - result: Dictionary of account information after successful login. + Dictionary of account information after successful login. """ # Before we actually log them in we check if they've already logged in diff --git a/synapse/rest/media/v1/media_repository.py b/synapse/rest/media/v1/media_repository.py index 328c0c5477..40b0d39eb2 100644 --- a/synapse/rest/media/v1/media_repository.py +++ b/synapse/rest/media/v1/media_repository.py @@ -344,8 +344,8 @@ class MediaRepository: download from remote server. Args: - server_name (str): Remote server_name where the media originated. - media_id (str): The media ID of the content (as defined by the + server_name: Remote server_name where the media originated. + media_id: The media ID of the content (as defined by the remote server). Returns: diff --git a/synapse/rest/media/v1/thumbnailer.py b/synapse/rest/media/v1/thumbnailer.py index 9b93b9b4f6..a48a4de92a 100644 --- a/synapse/rest/media/v1/thumbnailer.py +++ b/synapse/rest/media/v1/thumbnailer.py @@ -138,7 +138,7 @@ class Thumbnailer: """Rescales the image to the given dimensions. Returns: - BytesIO: the bytes of the encoded image ready to be written to disk + The bytes of the encoded image ready to be written to disk """ with self._resize(width, height) as scaled: return self._encode_image(scaled, output_type) @@ -155,7 +155,7 @@ class Thumbnailer: max_height: The largest possible height. Returns: - BytesIO: the bytes of the encoded image ready to be written to disk + The bytes of the encoded image ready to be written to disk """ if width * self.height > height * self.width: scaled_width = width diff --git a/synapse/server_notices/consent_server_notices.py b/synapse/server_notices/consent_server_notices.py index 698ca742ed..94025ba41f 100644 --- a/synapse/server_notices/consent_server_notices.py +++ b/synapse/server_notices/consent_server_notices.py @@ -113,9 +113,8 @@ def copy_with_str_subst(x: Any, substitutions: Any) -> Any: """Deep-copy a structure, carrying out string substitutions on any strings Args: - x (object): structure to be copied - substitutions (object): substitutions to be made - passed into the - string '%' operator + x: structure to be copied + substitutions: substitutions to be made - passed into the string '%' operator Returns: copy of x diff --git a/synapse/server_notices/resource_limits_server_notices.py b/synapse/server_notices/resource_limits_server_notices.py index 3134cd2d3d..a31a2c99a7 100644 --- a/synapse/server_notices/resource_limits_server_notices.py +++ b/synapse/server_notices/resource_limits_server_notices.py @@ -170,11 +170,13 @@ class ResourceLimitsServerNotices: room_id: The room id of the server notices room Returns: - bool: Is the room currently blocked - list: The list of pinned event IDs that are unrelated to limit blocking - This list can be used as a convenience in the case where the block - is to be lifted and the remaining pinned event references need to be - preserved + Tuple of: + Is the room currently blocked + + The list of pinned event IDs that are unrelated to limit blocking + This list can be used as a convenience in the case where the block + is to be lifted and the remaining pinned event references need to be + preserved """ currently_blocked = False pinned_state_event = None diff --git a/synapse/storage/controllers/persist_events.py b/synapse/storage/controllers/persist_events.py index 48976dc570..33ffef521b 100644 --- a/synapse/storage/controllers/persist_events.py +++ b/synapse/storage/controllers/persist_events.py @@ -204,9 +204,8 @@ class _EventPeristenceQueue(Generic[_PersistResult]): process to to so, calling the per_item_callback for each item. Args: - room_id (str): - task (_EventPersistQueueTask): A _PersistEventsTask or - _UpdateCurrentStateTask to process. + room_id: + task: A _PersistEventsTask or _UpdateCurrentStateTask to process. Returns: the result returned by the `_per_item_callback` passed to diff --git a/synapse/storage/databases/main/devices.py b/synapse/storage/databases/main/devices.py index aa58c2adc3..e114c733d1 100644 --- a/synapse/storage/databases/main/devices.py +++ b/synapse/storage/databases/main/devices.py @@ -535,7 +535,7 @@ class DeviceWorkerStore(RoomMemberWorkerStore, EndToEndKeyWorkerStore): limit: Maximum number of device updates to return Returns: - List: List of device update tuples: + List of device update tuples: - user_id - device_id - stream_id diff --git a/synapse/storage/databases/main/e2e_room_keys.py b/synapse/storage/databases/main/e2e_room_keys.py index af59be6b48..6240f9a75e 100644 --- a/synapse/storage/databases/main/e2e_room_keys.py +++ b/synapse/storage/databases/main/e2e_room_keys.py @@ -391,10 +391,10 @@ class EndToEndRoomKeyStore(SQLBaseStore): Returns: A dict giving the info metadata for this backup version, with fields including: - version(str) - algorithm(str) - auth_data(object): opaque dict supplied by the client - etag(int): tag of the keys in the backup + version (str) + algorithm (str) + auth_data (object): opaque dict supplied by the client + etag (int): tag of the keys in the backup """ def _get_e2e_room_keys_version_info_txn(txn: LoggingTransaction) -> JsonDict: diff --git a/synapse/storage/databases/main/end_to_end_keys.py b/synapse/storage/databases/main/end_to_end_keys.py index 2a4f58ed92..cf33e73e2b 100644 --- a/synapse/storage/databases/main/end_to_end_keys.py +++ b/synapse/storage/databases/main/end_to_end_keys.py @@ -412,10 +412,9 @@ class EndToEndKeyWorkerStore(EndToEndKeyBackgroundStore, CacheInvalidationWorker """Retrieve a number of one-time keys for a user Args: - user_id(str): id of user to get keys for - device_id(str): id of device to get keys for - key_ids(list[str]): list of key ids (excluding algorithm) to - retrieve + user_id: id of user to get keys for + device_id: id of device to get keys for + key_ids: list of key ids (excluding algorithm) to retrieve Returns: A map from (algorithm, key_id) to json string for key diff --git a/synapse/storage/databases/main/events.py b/synapse/storage/databases/main/events.py index c4acff5be6..d68f127f9b 100644 --- a/synapse/storage/databases/main/events.py +++ b/synapse/storage/databases/main/events.py @@ -1279,9 +1279,10 @@ class PersistEventsStore: Pick the earliest non-outlier if there is one, else the earliest one. Args: - events_and_contexts (list[(EventBase, EventContext)]): + events_and_contexts: + Returns: - list[(EventBase, EventContext)]: filtered list + filtered list """ new_events_and_contexts: OrderedDict[ str, Tuple[EventBase, EventContext] @@ -1307,9 +1308,8 @@ class PersistEventsStore: """Update min_depth for each room Args: - txn (twisted.enterprise.adbapi.Connection): db connection - events_and_contexts (list[(EventBase, EventContext)]): events - we are persisting + txn: db connection + events_and_contexts: events we are persisting """ depth_updates: Dict[str, int] = {} for event, context in events_and_contexts: @@ -1580,13 +1580,11 @@ class PersistEventsStore: """Update all the miscellaneous tables for new events Args: - txn (twisted.enterprise.adbapi.Connection): db connection - events_and_contexts (list[(EventBase, EventContext)]): events - we are persisting - all_events_and_contexts (list[(EventBase, EventContext)]): all - events that we were going to persist. This includes events - we've already persisted, etc, that wouldn't appear in - events_and_context. + txn: db connection + events_and_contexts: events we are persisting + all_events_and_contexts: all events that we were going to persist. + This includes events we've already persisted, etc, that wouldn't + appear in events_and_context. inhibit_local_membership_updates: Stop the local_current_membership from being updated by these events. This should be set to True for backfilled events because backfilled events in the past do diff --git a/synapse/storage/databases/main/events_worker.py b/synapse/storage/databases/main/events_worker.py index 467d20253d..8a104f7e93 100644 --- a/synapse/storage/databases/main/events_worker.py +++ b/synapse/storage/databases/main/events_worker.py @@ -1589,7 +1589,7 @@ class EventsWorkerStore(SQLBaseStore): room_id: The room ID to query. Returns: - dict[str:float] of complexity version to complexity. + Map of complexity version to complexity. """ state_events = await self.get_current_state_event_counts(room_id) diff --git a/synapse/storage/databases/main/monthly_active_users.py b/synapse/storage/databases/main/monthly_active_users.py index efd136a864..db9a24db5e 100644 --- a/synapse/storage/databases/main/monthly_active_users.py +++ b/synapse/storage/databases/main/monthly_active_users.py @@ -217,7 +217,7 @@ class MonthlyActiveUsersWorkerStore(RegistrationWorkerStore): def _reap_users(txn: LoggingTransaction, reserved_users: List[str]) -> None: """ Args: - reserved_users (tuple): reserved users to preserve + reserved_users: reserved users to preserve """ thirty_days_ago = int(self._clock.time_msec()) - (1000 * 60 * 60 * 24 * 30) @@ -370,8 +370,8 @@ class MonthlyActiveUsersWorkerStore(RegistrationWorkerStore): should not appear in the MAU stats). Args: - txn (cursor): - user_id (str): user to add/update + txn: + user_id: user to add/update """ assert ( self._update_on_this_worker @@ -401,7 +401,7 @@ class MonthlyActiveUsersWorkerStore(RegistrationWorkerStore): add the user to the monthly active tables Args: - user_id(str): the user_id to query + user_id: the user_id to query """ assert ( self._update_on_this_worker diff --git a/synapse/storage/databases/main/registration.py b/synapse/storage/databases/main/registration.py index 5167089e03..31f0f2bd3d 100644 --- a/synapse/storage/databases/main/registration.py +++ b/synapse/storage/databases/main/registration.py @@ -953,7 +953,7 @@ class RegistrationWorkerStore(CacheInvalidationWorkerStore): """Returns user id from threepid Args: - txn (cursor): + txn: medium: threepid medium e.g. email address: threepid address e.g. me@example.com @@ -1283,8 +1283,8 @@ class RegistrationWorkerStore(CacheInvalidationWorkerStore): """Sets an expiration date to the account with the given user ID. Args: - user_id (str): User ID to set an expiration date for. - use_delta (bool): If set to False, the expiration date for the user will be + user_id: User ID to set an expiration date for. + use_delta: If set to False, the expiration date for the user will be now + validity period. If set to True, this expiration date will be a random value in the [now + period - d ; now + period] range, d being a delta equal to 10% of the validity period. diff --git a/synapse/storage/databases/main/room.py b/synapse/storage/databases/main/room.py index 7d97f8f60e..4fbaefad73 100644 --- a/synapse/storage/databases/main/room.py +++ b/synapse/storage/databases/main/room.py @@ -2057,7 +2057,8 @@ class RoomStore(RoomBackgroundUpdateStore, RoomWorkerStore): Args: report_id: ID of reported event in database Returns: - event_report: json list of information from event report + JSON dict of information from an event report or None if the + report does not exist. """ def _get_event_report_txn( @@ -2130,8 +2131,9 @@ class RoomStore(RoomBackgroundUpdateStore, RoomWorkerStore): user_id: search for user_id. Ignored if user_id is None room_id: search for room_id. Ignored if room_id is None Returns: - event_reports: json list of event reports - count: total number of event reports matching the filter criteria + Tuple of: + json list of event reports + total number of event reports matching the filter criteria """ def _get_event_reports_paginate_txn( diff --git a/synapse/storage/databases/main/user_directory.py b/synapse/storage/databases/main/user_directory.py index ddb25b5cea..698d6f7515 100644 --- a/synapse/storage/databases/main/user_directory.py +++ b/synapse/storage/databases/main/user_directory.py @@ -185,9 +185,8 @@ class UserDirectoryBackgroundUpdateStore(StateDeltasStore): - who should be in the user_directory. Args: - progress (dict) - batch_size (int): Maximum number of state events to process - per cycle. + progress + batch_size: Maximum number of state events to process per cycle. Returns: number of events processed. @@ -708,10 +707,10 @@ class UserDirectoryStore(UserDirectoryBackgroundUpdateStore): Returns the rooms that a user is in. Args: - user_id(str): Must be a local user + user_id: Must be a local user Returns: - list: user_id + List of room IDs """ rows = await self.db_pool.simple_select_onecol( table="users_who_share_private_rooms", diff --git a/synapse/types.py b/synapse/types.py index 773f0438d5..f2d436ddc3 100644 --- a/synapse/types.py +++ b/synapse/types.py @@ -143,8 +143,8 @@ class Requester: Requester. Args: - store (DataStore): Used to convert AS ID to AS object - input (dict): A dict produced by `serialize` + store: Used to convert AS ID to AS object + input: A dict produced by `serialize` Returns: Requester diff --git a/synapse/util/async_helpers.py b/synapse/util/async_helpers.py index 7f1d41eb3c..d24c4f68c4 100644 --- a/synapse/util/async_helpers.py +++ b/synapse/util/async_helpers.py @@ -217,7 +217,8 @@ async def concurrently_execute( limit: Maximum number of conccurent executions. Returns: - Deferred: Resolved when all function invocations have finished. + None, when all function invocations have finished. The return values + from those functions are discarded. """ it = iter(args) diff --git a/synapse/util/caches/__init__.py b/synapse/util/caches/__init__.py index f7c3a6794e..9387632d0d 100644 --- a/synapse/util/caches/__init__.py +++ b/synapse/util/caches/__init__.py @@ -197,7 +197,7 @@ def register_cache( resize_callback: A function which can be called to resize the cache. Returns: - CacheMetric: an object which provides inc_{hits,misses,evictions} methods + an object which provides inc_{hits,misses,evictions} methods """ if resizable: if not resize_callback: diff --git a/synapse/util/caches/deferred_cache.py b/synapse/util/caches/deferred_cache.py index bcb1cba362..bf7bd351e0 100644 --- a/synapse/util/caches/deferred_cache.py +++ b/synapse/util/caches/deferred_cache.py @@ -153,7 +153,7 @@ class DeferredCache(Generic[KT, VT]): Args: key: callback: Gets called when the entry in the cache is invalidated - update_metrics (bool): whether to update the cache hit rate metrics + update_metrics: whether to update the cache hit rate metrics Returns: A Deferred which completes with the result. Note that this may later fail diff --git a/synapse/util/caches/dictionary_cache.py b/synapse/util/caches/dictionary_cache.py index fa91479c97..5eaf70c7ab 100644 --- a/synapse/util/caches/dictionary_cache.py +++ b/synapse/util/caches/dictionary_cache.py @@ -169,10 +169,11 @@ class DictionaryCache(Generic[KT, DKT, DV]): if it is in the cache. Returns: - DictionaryEntry: If `dict_keys` is not None then `DictionaryEntry` - will contain include the keys that are in the cache. If None then - will either return the full dict if in the cache, or the empty - dict (with `full` set to False) if it isn't. + If `dict_keys` is not None then `DictionaryEntry` will contain include + the keys that are in the cache. + + If None then will either return the full dict if in the cache, or the + empty dict (with `full` set to False) if it isn't. """ if dict_keys is None: # The caller wants the full set of dictionary keys for this cache key diff --git a/synapse/util/caches/expiringcache.py b/synapse/util/caches/expiringcache.py index c6a5d0dfc0..01ad02af67 100644 --- a/synapse/util/caches/expiringcache.py +++ b/synapse/util/caches/expiringcache.py @@ -207,7 +207,7 @@ class ExpiringCache(Generic[KT, VT]): items from the cache. Returns: - bool: Whether the cache changed size or not. + Whether the cache changed size or not. """ new_size = int(self._original_max_size * factor) if new_size != self._max_size: diff --git a/synapse/util/caches/lrucache.py b/synapse/util/caches/lrucache.py index aa93109d13..dcf0eac3bf 100644 --- a/synapse/util/caches/lrucache.py +++ b/synapse/util/caches/lrucache.py @@ -389,11 +389,11 @@ class LruCache(Generic[KT, VT]): cache_name: The name of this cache, for the prometheus metrics. If unset, no metrics will be reported on this cache. - cache_type (type): + cache_type: type of underlying cache to be used. Typically one of dict or TreeCache. - size_callback (func(V) -> int | None): + size_callback: metrics_collection_callback: metrics collection callback. This is called early in the metrics @@ -403,7 +403,7 @@ class LruCache(Generic[KT, VT]): Ignored if cache_name is None. - apply_cache_factor_from_config (bool): If true, `max_size` will be + apply_cache_factor_from_config: If true, `max_size` will be multiplied by a cache factor derived from the homeserver config clock: @@ -796,7 +796,7 @@ class LruCache(Generic[KT, VT]): items from the cache. Returns: - bool: Whether the cache changed size or not. + Whether the cache changed size or not. """ if not self.apply_cache_factor_from_config: return False diff --git a/synapse/util/ratelimitutils.py b/synapse/util/ratelimitutils.py index 9f64fed0d7..2aceb1a47f 100644 --- a/synapse/util/ratelimitutils.py +++ b/synapse/util/ratelimitutils.py @@ -183,7 +183,7 @@ class FederationRateLimiter: # Handle request ... Args: - host (str): Origin of incoming request. + host: Origin of incoming request. Returns: context manager which returns a deferred. diff --git a/synapse/util/threepids.py b/synapse/util/threepids.py index 1e9c2faa64..54bc7589fd 100644 --- a/synapse/util/threepids.py +++ b/synapse/util/threepids.py @@ -48,7 +48,7 @@ async def check_3pid_allowed( registration: whether we want to bind the 3PID as part of registering a new user. Returns: - bool: whether the 3PID medium/address is allowed to be added to this HS + whether the 3PID medium/address is allowed to be added to this HS """ if not await hs.get_password_auth_provider().is_3pid_allowed( medium, address, registration diff --git a/synapse/util/wheel_timer.py b/synapse/util/wheel_timer.py index 177e198e7e..b1ec7f4bd8 100644 --- a/synapse/util/wheel_timer.py +++ b/synapse/util/wheel_timer.py @@ -90,10 +90,10 @@ class WheelTimer(Generic[T]): """Fetch any objects that have timed out Args: - now (ms): Current time in msec + now: Current time in msec Returns: - list: List of objects that have timed out + List of objects that have timed out """ now_key = int(now / self.bucket_size) diff --git a/tests/http/__init__.py b/tests/http/__init__.py index e74f7f5b48..093537adef 100644 --- a/tests/http/__init__.py +++ b/tests/http/__init__.py @@ -13,6 +13,7 @@ # limitations under the License. import os.path import subprocess +from typing import List from zope.interface import implementer @@ -70,14 +71,14 @@ subjectAltName = %(sanentries)s """ -def create_test_cert_file(sanlist): +def create_test_cert_file(sanlist: List[bytes]) -> str: """build an x509 certificate file Args: - sanlist: list[bytes]: a list of subjectAltName values for the cert + sanlist: a list of subjectAltName values for the cert Returns: - str: the path to the file + The path to the file """ global cert_file_count csr_filename = "server.csr" diff --git a/tests/replication/slave/storage/test_events.py b/tests/replication/slave/storage/test_events.py index 96f3880923..dce71f7334 100644 --- a/tests/replication/slave/storage/test_events.py +++ b/tests/replication/slave/storage/test_events.py @@ -143,6 +143,7 @@ class EventsWorkerStoreTestCase(BaseSlavedStoreTestCase): self.persist(type="m.room.create", key="", creator=USER_ID) self.check("get_invited_rooms_for_local_user", [USER_ID_2], []) event = self.persist(type="m.room.member", key=USER_ID_2, membership="invite") + assert event.internal_metadata.stream_ordering is not None self.replicate() @@ -230,6 +231,7 @@ class EventsWorkerStoreTestCase(BaseSlavedStoreTestCase): j2 = self.persist( type="m.room.member", sender=USER_ID_2, key=USER_ID_2, membership="join" ) + assert j2.internal_metadata.stream_ordering is not None self.replicate() expected_pos = PersistedEventPosition( @@ -287,6 +289,7 @@ class EventsWorkerStoreTestCase(BaseSlavedStoreTestCase): ) ) self.replicate() + assert j2.internal_metadata.stream_ordering is not None event_source = RoomEventSource(self.hs) event_source.store = self.slaved_store @@ -336,10 +339,10 @@ class EventsWorkerStoreTestCase(BaseSlavedStoreTestCase): event_id = 0 - def persist(self, backfill=False, **kwargs): + def persist(self, backfill=False, **kwargs) -> FrozenEvent: """ Returns: - synapse.events.FrozenEvent: The event that was persisted. + The event that was persisted. """ event, context = self.build_event(**kwargs) diff --git a/tests/replication/test_multi_media_repo.py b/tests/replication/test_multi_media_repo.py index 13aa5eb51a..96cdf2c45b 100644 --- a/tests/replication/test_multi_media_repo.py +++ b/tests/replication/test_multi_media_repo.py @@ -15,8 +15,9 @@ import logging import os from typing import Optional, Tuple +from twisted.internet.interfaces import IOpenSSLServerConnectionCreator from twisted.internet.protocol import Factory -from twisted.protocols.tls import TLSMemoryBIOFactory +from twisted.protocols.tls import TLSMemoryBIOFactory, TLSMemoryBIOProtocol from twisted.web.http import HTTPChannel from twisted.web.server import Request @@ -102,7 +103,7 @@ class MediaRepoShardTestCase(BaseMultiWorkerStreamTestCase): ) # fish the test server back out of the server-side TLS protocol. - http_server = server_tls_protocol.wrappedProtocol + http_server: HTTPChannel = server_tls_protocol.wrappedProtocol # type: ignore[assignment] # give the reactor a pump to get the TLS juices flowing. self.reactor.pump((0.1,)) @@ -238,16 +239,15 @@ def get_connection_factory(): return test_server_connection_factory -def _build_test_server(connection_creator): +def _build_test_server( + connection_creator: IOpenSSLServerConnectionCreator, +) -> TLSMemoryBIOProtocol: """Construct a test server This builds an HTTP channel, wrapped with a TLSMemoryBIOProtocol Args: - connection_creator (IOpenSSLServerConnectionCreator): thing to build - SSL connections - sanlist (list[bytes]): list of the SAN entries for the cert returned - by the server + connection_creator: thing to build SSL connections Returns: TLSMemoryBIOProtocol diff --git a/tests/server_notices/test_resource_limits_server_notices.py b/tests/server_notices/test_resource_limits_server_notices.py index bf403045e9..7cbc40736c 100644 --- a/tests/server_notices/test_resource_limits_server_notices.py +++ b/tests/server_notices/test_resource_limits_server_notices.py @@ -11,6 +11,7 @@ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. +from typing import Tuple from unittest.mock import Mock from twisted.test.proto_helpers import MemoryReactor @@ -350,14 +351,15 @@ class TestResourceLimitsServerNoticesWithRealRooms(unittest.HomeserverTestCase): self.assertTrue(notice_in_room, "No server notice in room") - def _trigger_notice_and_join(self): + def _trigger_notice_and_join(self) -> Tuple[str, str, str]: """Creates enough active users to hit the MAU limit and trigger a system notice about it, then joins the system notices room with one of the users created. Returns: - user_id (str): The ID of the user that joined the room. - tok (str): The access token of the user that joined the room. - room_id (str): The ID of the room that's been joined. + A tuple of: + user_id: The ID of the user that joined the room. + tok: The access token of the user that joined the room. + room_id: The ID of the room that's been joined. """ user_id = None tok = None diff --git a/tests/unittest.py b/tests/unittest.py index 5116be338e..a120c2976c 100644 --- a/tests/unittest.py +++ b/tests/unittest.py @@ -360,13 +360,13 @@ class HomeserverTestCase(TestCase): store.db_pool.updates.do_next_background_update(False), by=0.1 ) - def make_homeserver(self, reactor, clock): + def make_homeserver(self, reactor: MemoryReactor, clock: Clock): """ Make and return a homeserver. Args: reactor: A Twisted Reactor, or something that pretends to be one. - clock (synapse.util.Clock): The Clock, associated with the reactor. + clock: The Clock, associated with the reactor. Returns: A homeserver suitable for testing. @@ -426,9 +426,8 @@ class HomeserverTestCase(TestCase): Args: reactor: A Twisted Reactor, or something that pretends to be one. - clock (synapse.util.Clock): The Clock, associated with the reactor. - homeserver (synapse.server.HomeServer): The HomeServer to test - against. + clock: The Clock, associated with the reactor. + homeserver: The HomeServer to test against. Function to optionally be overridden in subclasses. """ @@ -452,11 +451,10 @@ class HomeserverTestCase(TestCase): given content. Args: - method (bytes/unicode): The HTTP request method ("verb"). - path (bytes/unicode): The HTTP path, suitably URL encoded (e.g. - escaped UTF-8 & spaces and such). - content (bytes or dict): The body of the request. JSON-encoded, if - a dict. + method: The HTTP request method ("verb"). + path: The HTTP path, suitably URL encoded (e.g. escaped UTF-8 & spaces + and such). content (bytes or dict): The body of the request. + JSON-encoded, if a dict. shorthand: Whether to try and be helpful and prefix the given URL with the usual REST API path, if it doesn't contain it. federation_auth_origin: if set to not-None, we will add a fake -- cgit 1.5.1 From 1799a54a545618782840a60950ef4b64da9ee24d Mon Sep 17 00:00:00 2001 From: Patrick Cloke Date: Tue, 22 Nov 2022 07:26:11 -0500 Subject: Batch fetch bundled annotations (#14491) Avoid an n+1 query problem and fetch the bundled aggregations for m.annotation relations in a single query instead of a query per event. This applies similar logic for as was previously done for edits in 8b309adb436c162510ed1402f33b8741d71fc058 (#11660) and threads in b65acead428653b988351ae8d7b22127a22039cd (#11752). --- changelog.d/14491.feature | 1 + synapse/handlers/relations.py | 197 ++++++++++++++++------------ synapse/storage/databases/main/relations.py | 139 ++++++++++++-------- synapse/util/caches/descriptors.py | 2 +- tests/rest/client/test_relations.py | 4 +- 5 files changed, 202 insertions(+), 141 deletions(-) create mode 100644 changelog.d/14491.feature (limited to 'synapse/util') diff --git a/changelog.d/14491.feature b/changelog.d/14491.feature new file mode 100644 index 0000000000..4fca7282f7 --- /dev/null +++ b/changelog.d/14491.feature @@ -0,0 +1 @@ +Reduce database load of [Client-Server endpoints](https://spec.matrix.org/v1.4/client-server-api/#aggregations) which return bundled aggregations. diff --git a/synapse/handlers/relations.py b/synapse/handlers/relations.py index 8e71dda970..ca94239f61 100644 --- a/synapse/handlers/relations.py +++ b/synapse/handlers/relations.py @@ -13,7 +13,16 @@ # limitations under the License. import enum import logging -from typing import TYPE_CHECKING, Dict, FrozenSet, Iterable, List, Optional, Tuple +from typing import ( + TYPE_CHECKING, + Collection, + Dict, + FrozenSet, + Iterable, + List, + Optional, + Tuple, +) import attr @@ -259,48 +268,64 @@ class RelationsHandler: e.msg, ) - async def get_annotations_for_event( - self, - event_id: str, - room_id: str, - limit: int = 5, - ignored_users: FrozenSet[str] = frozenset(), - ) -> List[JsonDict]: - """Get a list of annotations on the event, grouped by event type and + async def get_annotations_for_events( + self, event_ids: Collection[str], ignored_users: FrozenSet[str] = frozenset() + ) -> Dict[str, List[JsonDict]]: + """Get a list of annotations to the given events, grouped by event type and aggregation key, sorted by count. - This is used e.g. to get the what and how many reactions have happend + This is used e.g. to get the what and how many reactions have happened on an event. Args: - event_id: Fetch events that relate to this event ID. - room_id: The room the event belongs to. - limit: Only fetch the `limit` groups. + event_ids: Fetch events that relate to these event IDs. ignored_users: The users ignored by the requesting user. Returns: - List of groups of annotations that match. Each row is a dict with - `type`, `key` and `count` fields. + A map of event IDs to a list of groups of annotations that match. + Each entry is a dict with `type`, `key` and `count` fields. """ # Get the base results for all users. - full_results = await self._main_store.get_aggregation_groups_for_event( - event_id, room_id, limit + full_results = await self._main_store.get_aggregation_groups_for_events( + event_ids ) + # Avoid additional logic if there are no ignored users. + if not ignored_users: + return { + event_id: results + for event_id, results in full_results.items() + if results + } + # Then subtract off the results for any ignored users. ignored_results = await self._main_store.get_aggregation_groups_for_users( - event_id, room_id, limit, ignored_users + [event_id for event_id, results in full_results.items() if results], + ignored_users, ) - filtered_results = [] - for result in full_results: - key = (result["type"], result["key"]) - if key in ignored_results: - result = result.copy() - result["count"] -= ignored_results[key] - if result["count"] <= 0: - continue - filtered_results.append(result) + filtered_results = {} + for event_id, results in full_results.items(): + # If no annotations, skip. + if not results: + continue + + # If there are not ignored results for this event, copy verbatim. + if event_id not in ignored_results: + filtered_results[event_id] = results + continue + + # Otherwise, subtract out the ignored results. + event_ignored_results = ignored_results[event_id] + for result in results: + key = (result["type"], result["key"]) + if key in event_ignored_results: + # Ensure to not modify the cache. + result = result.copy() + result["count"] -= event_ignored_results[key] + if result["count"] <= 0: + continue + filtered_results.setdefault(event_id, []).append(result) return filtered_results @@ -366,59 +391,62 @@ class RelationsHandler: results = {} for event_id, summary in summaries.items(): - if summary: - thread_count, latest_thread_event = summary - - # Subtract off the count of any ignored users. - for ignored_user in ignored_users: - thread_count -= ignored_results.get((event_id, ignored_user), 0) - - # This is gnarly, but if the latest event is from an ignored user, - # attempt to find one that isn't from an ignored user. - if latest_thread_event.sender in ignored_users: - room_id = latest_thread_event.room_id - - # If the root event is not found, something went wrong, do - # not include a summary of the thread. - event = await self._event_handler.get_event(user, room_id, event_id) - if event is None: - continue + # If no thread, skip. + if not summary: + continue - potential_events, _ = await self.get_relations_for_event( - event_id, - event, - room_id, - RelationTypes.THREAD, - ignored_users, - ) + thread_count, latest_thread_event = summary - # If all found events are from ignored users, do not include - # a summary of the thread. - if not potential_events: - continue + # Subtract off the count of any ignored users. + for ignored_user in ignored_users: + thread_count -= ignored_results.get((event_id, ignored_user), 0) - # The *last* event returned is the one that is cared about. - event = await self._event_handler.get_event( - user, room_id, potential_events[-1].event_id - ) - # It is unexpected that the event will not exist. - if event is None: - logger.warning( - "Unable to fetch latest event in a thread with event ID: %s", - potential_events[-1].event_id, - ) - continue - latest_thread_event = event - - results[event_id] = _ThreadAggregation( - latest_event=latest_thread_event, - count=thread_count, - # If there's a thread summary it must also exist in the - # participated dictionary. - current_user_participated=events_by_id[event_id].sender == user_id - or participated[event_id], + # This is gnarly, but if the latest event is from an ignored user, + # attempt to find one that isn't from an ignored user. + if latest_thread_event.sender in ignored_users: + room_id = latest_thread_event.room_id + + # If the root event is not found, something went wrong, do + # not include a summary of the thread. + event = await self._event_handler.get_event(user, room_id, event_id) + if event is None: + continue + + potential_events, _ = await self.get_relations_for_event( + event_id, + event, + room_id, + RelationTypes.THREAD, + ignored_users, ) + # If all found events are from ignored users, do not include + # a summary of the thread. + if not potential_events: + continue + + # The *last* event returned is the one that is cared about. + event = await self._event_handler.get_event( + user, room_id, potential_events[-1].event_id + ) + # It is unexpected that the event will not exist. + if event is None: + logger.warning( + "Unable to fetch latest event in a thread with event ID: %s", + potential_events[-1].event_id, + ) + continue + latest_thread_event = event + + results[event_id] = _ThreadAggregation( + latest_event=latest_thread_event, + count=thread_count, + # If there's a thread summary it must also exist in the + # participated dictionary. + current_user_participated=events_by_id[event_id].sender == user_id + or participated[event_id], + ) + return results @trace @@ -496,17 +524,18 @@ class RelationsHandler: # (as that is what makes it part of the thread). relations_by_id[latest_thread_event.event_id] = RelationTypes.THREAD - # Fetch other relations per event. - for event in events_by_id.values(): - # Fetch any annotations (ie, reactions) to bundle with this event. - annotations = await self.get_annotations_for_event( - event.event_id, event.room_id, ignored_users=ignored_users - ) + # Fetch any annotations (ie, reactions) to bundle with this event. + annotations_by_event_id = await self.get_annotations_for_events( + events_by_id.keys(), ignored_users=ignored_users + ) + for event_id, annotations in annotations_by_event_id.items(): if annotations: - results.setdefault( - event.event_id, BundledAggregations() - ).annotations = {"chunk": annotations} + results.setdefault(event_id, BundledAggregations()).annotations = { + "chunk": annotations + } + # Fetch other relations per event. + for event in events_by_id.values(): # Fetch any references to bundle with this event. references, next_token = await self.get_relations_for_event( event.event_id, diff --git a/synapse/storage/databases/main/relations.py b/synapse/storage/databases/main/relations.py index ca431002c8..f96a16956a 100644 --- a/synapse/storage/databases/main/relations.py +++ b/synapse/storage/databases/main/relations.py @@ -20,6 +20,7 @@ from typing import ( FrozenSet, Iterable, List, + Mapping, Optional, Set, Tuple, @@ -394,106 +395,136 @@ class RelationsWorkerStore(SQLBaseStore): ) return result is not None - @cached(tree=True) - async def get_aggregation_groups_for_event( - self, event_id: str, room_id: str, limit: int = 5 - ) -> List[JsonDict]: - """Get a list of annotations on the event, grouped by event type and + @cached() + async def get_aggregation_groups_for_event(self, event_id: str) -> List[JsonDict]: + raise NotImplementedError() + + @cachedList( + cached_method_name="get_aggregation_groups_for_event", list_name="event_ids" + ) + async def get_aggregation_groups_for_events( + self, event_ids: Collection[str] + ) -> Mapping[str, Optional[List[JsonDict]]]: + """Get a list of annotations on the given events, grouped by event type and aggregation key, sorted by count. This is used e.g. to get the what and how many reactions have happend on an event. Args: - event_id: Fetch events that relate to this event ID. - room_id: The room the event belongs to. - limit: Only fetch the `limit` groups. + event_ids: Fetch events that relate to these event IDs. Returns: - List of groups of annotations that match. Each row is a dict with - `type`, `key` and `count` fields. + A map of event IDs to a list of groups of annotations that match. + Each entry is a dict with `type`, `key` and `count` fields. """ + # The number of entries to return per event ID. + limit = 5 - args = [ - event_id, - room_id, - RelationTypes.ANNOTATION, - limit, - ] + clause, args = make_in_list_sql_clause( + self.database_engine, "relates_to_id", event_ids + ) + args.append(RelationTypes.ANNOTATION) - sql = """ - SELECT type, aggregation_key, COUNT(DISTINCT sender) - FROM event_relations - INNER JOIN events USING (event_id) - WHERE relates_to_id = ? AND room_id = ? AND relation_type = ? - GROUP BY relation_type, type, aggregation_key - ORDER BY COUNT(*) DESC - LIMIT ? + sql = f""" + SELECT + relates_to_id, + annotation.type, + aggregation_key, + COUNT(DISTINCT annotation.sender) + FROM events AS annotation + INNER JOIN event_relations USING (event_id) + INNER JOIN events AS parent ON + parent.event_id = relates_to_id + AND parent.room_id = annotation.room_id + WHERE + {clause} + AND relation_type = ? + GROUP BY relates_to_id, annotation.type, aggregation_key + ORDER BY relates_to_id, COUNT(*) DESC """ - def _get_aggregation_groups_for_event_txn( + def _get_aggregation_groups_for_events_txn( txn: LoggingTransaction, - ) -> List[JsonDict]: + ) -> Mapping[str, List[JsonDict]]: txn.execute(sql, args) - return [{"type": row[0], "key": row[1], "count": row[2]} for row in txn] + result: Dict[str, List[JsonDict]] = {} + for event_id, type, key, count in cast( + List[Tuple[str, str, str, int]], txn + ): + event_results = result.setdefault(event_id, []) + + # Limit the number of results per event ID. + if len(event_results) == limit: + continue + + event_results.append({"type": type, "key": key, "count": count}) + + return result return await self.db_pool.runInteraction( - "get_aggregation_groups_for_event", _get_aggregation_groups_for_event_txn + "get_aggregation_groups_for_events", _get_aggregation_groups_for_events_txn ) async def get_aggregation_groups_for_users( - self, - event_id: str, - room_id: str, - limit: int, - users: FrozenSet[str] = frozenset(), - ) -> Dict[Tuple[str, str], int]: + self, event_ids: Collection[str], users: FrozenSet[str] + ) -> Dict[str, Dict[Tuple[str, str], int]]: """Fetch the partial aggregations for an event for specific users. This is used, in conjunction with get_aggregation_groups_for_event, to remove information from the results for ignored users. Args: - event_id: Fetch events that relate to this event ID. - room_id: The room the event belongs to. - limit: Only fetch the `limit` groups. + event_ids: Fetch events that relate to these event IDs. users: The users to fetch information for. Returns: - A map of (event type, aggregation key) to a count of users. + A map of event ID to a map of (event type, aggregation key) to a + count of users. """ if not users: return {} - args: List[Union[str, int]] = [ - event_id, - room_id, - RelationTypes.ANNOTATION, - ] + events_sql, args = make_in_list_sql_clause( + self.database_engine, "relates_to_id", event_ids + ) users_sql, users_args = make_in_list_sql_clause( - self.database_engine, "sender", users + self.database_engine, "annotation.sender", users ) args.extend(users_args) + args.append(RelationTypes.ANNOTATION) sql = f""" - SELECT type, aggregation_key, COUNT(DISTINCT sender) - FROM event_relations - INNER JOIN events USING (event_id) - WHERE relates_to_id = ? AND room_id = ? AND relation_type = ? AND {users_sql} - GROUP BY relation_type, type, aggregation_key - ORDER BY COUNT(*) DESC - LIMIT ? + SELECT + relates_to_id, + annotation.type, + aggregation_key, + COUNT(DISTINCT annotation.sender) + FROM events AS annotation + INNER JOIN event_relations USING (event_id) + INNER JOIN events AS parent ON + parent.event_id = relates_to_id + AND parent.room_id = annotation.room_id + WHERE {events_sql} AND {users_sql} AND relation_type = ? + GROUP BY relates_to_id, annotation.type, aggregation_key + ORDER BY relates_to_id, COUNT(*) DESC """ def _get_aggregation_groups_for_users_txn( txn: LoggingTransaction, - ) -> Dict[Tuple[str, str], int]: - txn.execute(sql, args + [limit]) + ) -> Dict[str, Dict[Tuple[str, str], int]]: + txn.execute(sql, args) - return {(row[0], row[1]): row[2] for row in txn} + result: Dict[str, Dict[Tuple[str, str], int]] = {} + for event_id, type, key, count in cast( + List[Tuple[str, str, str, int]], txn + ): + result.setdefault(event_id, {})[(type, key)] = count + + return result return await self.db_pool.runInteraction( "get_aggregation_groups_for_users", _get_aggregation_groups_for_users_txn diff --git a/synapse/util/caches/descriptors.py b/synapse/util/caches/descriptors.py index 75428d19ba..72227359b9 100644 --- a/synapse/util/caches/descriptors.py +++ b/synapse/util/caches/descriptors.py @@ -503,7 +503,7 @@ def cachedList( is specified as a list that is iterated through to lookup keys in the original cache. A new tuple consisting of the (deduplicated) keys that weren't in the cache gets passed to the original function, which is expected to results - in a map of key to value for each passed value. THe new results are stored in the + in a map of key to value for each passed value. The new results are stored in the original cache. Note that any missing values are cached as None. Args: diff --git a/tests/rest/client/test_relations.py b/tests/rest/client/test_relations.py index e3d801f7a8..2d2b683548 100644 --- a/tests/rest/client/test_relations.py +++ b/tests/rest/client/test_relations.py @@ -1108,7 +1108,7 @@ class BundledAggregationsTestCase(BaseRelationsTestCase): # The "user" sent the root event and is making queries for the bundled # aggregations: they have participated. - self._test_bundled_aggregations(RelationTypes.THREAD, _gen_assert(True), 9) + self._test_bundled_aggregations(RelationTypes.THREAD, _gen_assert(True), 8) # The "user2" sent replies in the thread and is making queries for the # bundled aggregations: they have participated. # @@ -1170,7 +1170,7 @@ class BundledAggregationsTestCase(BaseRelationsTestCase): bundled_aggregations["latest_event"].get("unsigned"), ) - self._test_bundled_aggregations(RelationTypes.THREAD, assert_thread, 9) + self._test_bundled_aggregations(RelationTypes.THREAD, assert_thread, 8) def test_nested_thread(self) -> None: """ -- cgit 1.5.1