diff options
Diffstat (limited to 'synapse')
| -rw-r--r-- | synapse/appservice/__init__.py | 37 | ||||
| -rw-r--r-- | synapse/config/appservice.py | 1 | ||||
| -rw-r--r-- | synapse/config/push.py | 44 | ||||
| -rw-r--r-- | synapse/federation/federation_client.py | 4 | ||||
| -rw-r--r-- | synapse/handlers/appservice.py | 4 | ||||
| -rw-r--r-- | synapse/handlers/groups_local.py | 11 | ||||
| -rw-r--r-- | synapse/handlers/initial_sync.py | 4 | ||||
| -rw-r--r-- | synapse/handlers/presence.py | 2 | ||||
| -rw-r--r-- | synapse/handlers/room_list.py | 50 | ||||
| -rw-r--r-- | synapse/push/httppusher.py | 3 | ||||
| -rw-r--r-- | synapse/push/pusherpool.py | 6 | ||||
| -rw-r--r-- | synapse/replication/slave/storage/events.py | 39 | ||||
| -rw-r--r-- | synapse/rest/client/versions.py | 1 | ||||
| -rw-r--r-- | synapse/storage/_base.py | 74 | ||||
| -rw-r--r-- | synapse/storage/pusher.py | 55 | ||||
| -rw-r--r-- | synapse/storage/state.py | 441 | ||||
| -rw-r--r-- | synapse/storage/stream.py | 4 | ||||
| -rw-r--r-- | synapse/util/async.py | 6 | ||||
| -rw-r--r-- | synapse/util/distributor.py | 22 | ||||
| -rw-r--r-- | synapse/util/logcontext.py | 61 | ||||
| -rw-r--r-- | synapse/visibility.py | 4 |
21 files changed, 446 insertions, 427 deletions
diff --git a/synapse/appservice/__init__.py b/synapse/appservice/__init__.py index b989007314..d5a7a5ce2f 100644 --- a/synapse/appservice/__init__.py +++ b/synapse/appservice/__init__.py @@ -14,6 +14,7 @@ # limitations under the License. from synapse.api.constants import EventTypes from synapse.util.caches.descriptors import cachedInlineCallbacks +from synapse.types import GroupID, get_domain_from_id from twisted.internet import defer @@ -81,12 +82,13 @@ class ApplicationService(object): # values. NS_LIST = [NS_USERS, NS_ALIASES, NS_ROOMS] - def __init__(self, token, url=None, namespaces=None, hs_token=None, + def __init__(self, token, hostname, url=None, namespaces=None, hs_token=None, sender=None, id=None, protocols=None, rate_limited=True): self.token = token self.url = url self.hs_token = hs_token self.sender = sender + self.server_name = hostname self.namespaces = self._check_namespaces(namespaces) self.id = id @@ -125,6 +127,24 @@ class ApplicationService(object): raise ValueError( "Expected bool for 'exclusive' in ns '%s'" % ns ) + group_id = regex_obj.get("group_id") + if group_id: + if not isinstance(group_id, str): + raise ValueError( + "Expected string for 'group_id' in ns '%s'" % ns + ) + try: + GroupID.from_string(group_id) + except Exception: + raise ValueError( + "Expected valid group ID for 'group_id' in ns '%s'" % ns + ) + + if get_domain_from_id(group_id) != self.server_name: + raise ValueError( + "Expected 'group_id' to be this host in ns '%s'" % ns + ) + regex = regex_obj.get("regex") if isinstance(regex, basestring): regex_obj["regex"] = re.compile(regex) # Pre-compile regex @@ -251,6 +271,21 @@ class ApplicationService(object): if regex_obj["exclusive"] ] + def get_groups_for_user(self, user_id): + """Get the groups that this user is associated with by this AS + + Args: + user_id (str): The ID of the user. + + Returns: + iterable[str]: an iterable that yields group_id strings. + """ + return ( + regex_obj["group_id"] + for regex_obj in self.namespaces[ApplicationService.NS_USERS] + if "group_id" in regex_obj and regex_obj["regex"].match(user_id) + ) + def is_rate_limited(self): return self.rate_limited diff --git a/synapse/config/appservice.py b/synapse/config/appservice.py index 82c50b8240..aba0aec6e8 100644 --- a/synapse/config/appservice.py +++ b/synapse/config/appservice.py @@ -154,6 +154,7 @@ def _load_appservice(hostname, as_info, config_filename): ) return ApplicationService( token=as_info["as_token"], + hostname=hostname, url=as_info["url"], namespaces=as_info["namespaces"], hs_token=as_info["hs_token"], diff --git a/synapse/config/push.py b/synapse/config/push.py index 9c68318b40..b7e0d46afa 100644 --- a/synapse/config/push.py +++ b/synapse/config/push.py @@ -1,5 +1,6 @@ # -*- coding: utf-8 -*- # Copyright 2015, 2016 OpenMarket Ltd +# Copyright 2017 New Vector Ltd # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. @@ -18,28 +19,43 @@ from ._base import Config class PushConfig(Config): def read_config(self, config): - self.push_redact_content = False + push_config = config.get("push", {}) + self.push_include_content = push_config.get("include_content", True) + # There was a a 'redact_content' setting but mistakenly read from the + # 'email'section'. Check for the flag in the 'push' section, and log, + # but do not honour it to avoid nasty surprises when people upgrade. + if push_config.get("redact_content") is not None: + print( + "The push.redact_content content option has never worked. " + "Please set push.include_content if you want this behaviour" + ) + + # Now check for the one in the 'email' section and honour it, + # with a warning. push_config = config.get("email", {}) - self.push_redact_content = push_config.get("redact_content", False) + redact_content = push_config.get("redact_content") + if redact_content is not None: + print( + "The 'email.redact_content' option is deprecated: " + "please set push.include_content instead" + ) + self.push_include_content = not redact_content def default_config(self, config_dir_path, server_name, **kwargs): return """ - # Control how push messages are sent to google/apple to notifications. - # Normally every message said in a room with one or more people using - # mobile devices will be posted to a push server hosted by matrix.org - # which is registered with google and apple in order to allow push - # notifications to be sent to these mobile devices. - # - # Setting redact_content to true will make the push messages contain no - # message content which will provide increased privacy. This is a - # temporary solution pending improvements to Android and iPhone apps - # to get content from the app rather than the notification. - # + # Clients requesting push notifications can either have the body of + # the message sent in the notification poke along with other details + # like the sender, or just the event ID and room ID (`event_id_only`). + # If clients choose the former, this option controls whether the + # notification request includes the content of the event (other details + # like the sender are still included). For `event_id_only` push, it + # has no effect. + # For modern android devices the notification content will still appear # because it is loaded by the app. iPhone, however will send a # notification saying only that a message arrived and who it came from. # #push: - # redact_content: false + # include_content: true """ diff --git a/synapse/federation/federation_client.py b/synapse/federation/federation_client.py index 7c5e5d957f..b8f02f5391 100644 --- a/synapse/federation/federation_client.py +++ b/synapse/federation/federation_client.py @@ -25,7 +25,7 @@ from synapse.api.errors import ( from synapse.util import unwrapFirstError, logcontext from synapse.util.caches.expiringcache import ExpiringCache from synapse.util.logutils import log_function -from synapse.util.logcontext import preserve_fn, preserve_context_over_deferred +from synapse.util.logcontext import make_deferred_yieldable, preserve_fn from synapse.events import FrozenEvent, builder import synapse.metrics @@ -420,7 +420,7 @@ class FederationClient(FederationBase): for e_id in batch ] - res = yield preserve_context_over_deferred( + res = yield make_deferred_yieldable( defer.DeferredList(deferreds, consumeErrors=True) ) for success, result in res: diff --git a/synapse/handlers/appservice.py b/synapse/handlers/appservice.py index 543bf28aec..feca3e4c10 100644 --- a/synapse/handlers/appservice.py +++ b/synapse/handlers/appservice.py @@ -17,7 +17,7 @@ from twisted.internet import defer from synapse.api.constants import EventTypes from synapse.util.metrics import Measure -from synapse.util.logcontext import preserve_fn, preserve_context_over_deferred +from synapse.util.logcontext import make_deferred_yieldable, preserve_fn import logging @@ -159,7 +159,7 @@ class ApplicationServicesHandler(object): def query_3pe(self, kind, protocol, fields): services = yield self._get_services_for_3pn(protocol) - results = yield preserve_context_over_deferred(defer.DeferredList([ + results = yield make_deferred_yieldable(defer.DeferredList([ preserve_fn(self.appservice_api.query_3pe)(service, kind, protocol, fields) for service in services ], consumeErrors=True)) diff --git a/synapse/handlers/groups_local.py b/synapse/handlers/groups_local.py index da00aeb0f4..7e5d3f148d 100644 --- a/synapse/handlers/groups_local.py +++ b/synapse/handlers/groups_local.py @@ -375,6 +375,12 @@ class GroupsLocalHandler(object): def get_publicised_groups_for_user(self, user_id): if self.hs.is_mine_id(user_id): result = yield self.store.get_publicised_groups_for_user(user_id) + + # Check AS associated groups for this user - this depends on the + # RegExps in the AS registration file (under `users`) + for app_service in self.store.get_app_services(): + result.extend(app_service.get_groups_for_user(user_id)) + defer.returnValue({"groups": result}) else: result = yield self.transport_client.get_publicised_groups_for_user( @@ -415,4 +421,9 @@ class GroupsLocalHandler(object): uid ) + # Check AS associated groups for this user - this depends on the + # RegExps in the AS registration file (under `users`) + for app_service in self.store.get_app_services(): + results[uid].extend(app_service.get_groups_for_user(uid)) + defer.returnValue({"users": results}) diff --git a/synapse/handlers/initial_sync.py b/synapse/handlers/initial_sync.py index 9718d4abc5..c5267b4b84 100644 --- a/synapse/handlers/initial_sync.py +++ b/synapse/handlers/initial_sync.py @@ -27,7 +27,7 @@ from synapse.types import ( from synapse.util import unwrapFirstError from synapse.util.async import concurrently_execute from synapse.util.caches.snapshot_cache import SnapshotCache -from synapse.util.logcontext import preserve_fn, preserve_context_over_deferred +from synapse.util.logcontext import make_deferred_yieldable, preserve_fn from synapse.visibility import filter_events_for_client from ._base import BaseHandler @@ -163,7 +163,7 @@ class InitialSyncHandler(BaseHandler): lambda states: states[event.event_id] ) - (messages, token), current_state = yield preserve_context_over_deferred( + (messages, token), current_state = yield make_deferred_yieldable( defer.gatherResults( [ preserve_fn(self.store.get_recent_events_for_room)( diff --git a/synapse/handlers/presence.py b/synapse/handlers/presence.py index fa96ea69cd..cb158ba962 100644 --- a/synapse/handlers/presence.py +++ b/synapse/handlers/presence.py @@ -1199,7 +1199,7 @@ def handle_timeout(state, is_mine, syncing_user_ids, now): ) changed = True else: - # We expect to be poked occaisonally by the other side. + # We expect to be poked occasionally by the other side. # This is to protect against forgetful/buggy servers, so that # no one gets stuck online forever. if now - state.last_federation_update_ts > FEDERATION_TIMEOUT: diff --git a/synapse/handlers/room_list.py b/synapse/handlers/room_list.py index 2cf34e51cb..bb40075387 100644 --- a/synapse/handlers/room_list.py +++ b/synapse/handlers/room_list.py @@ -154,6 +154,8 @@ class RoomListHandler(BaseHandler): # We want larger rooms to be first, hence negating num_joined_users rooms_to_order_value[room_id] = (-num_joined_users, room_id) + logger.info("Getting ordering for %i rooms since %s", + len(room_ids), stream_token) yield concurrently_execute(get_order_for_room, room_ids, 10) sorted_entries = sorted(rooms_to_order_value.items(), key=lambda e: e[1]) @@ -181,34 +183,42 @@ class RoomListHandler(BaseHandler): rooms_to_scan = rooms_to_scan[:since_token.current_limit] rooms_to_scan.reverse() - # Actually generate the entries. _append_room_entry_to_chunk will append to - # chunk but will stop if len(chunk) > limit - chunk = [] - if limit and not search_filter: + logger.info("After sorting and filtering, %i rooms remain", + len(rooms_to_scan)) + + # _append_room_entry_to_chunk will append to chunk but will stop if + # len(chunk) > limit + # + # Normally we will generate enough results on the first iteration here, + # but if there is a search filter, _append_room_entry_to_chunk may + # filter some results out, in which case we loop again. + # + # We don't want to scan over the entire range either as that + # would potentially waste a lot of work. + # + # XXX if there is no limit, we may end up DoSing the server with + # calls to get_current_state_ids for every single room on the + # server. Surely we should cap this somehow? + # + if limit: step = limit + 1 - for i in xrange(0, len(rooms_to_scan), step): - # We iterate here because the vast majority of cases we'll stop - # at first iteration, but occaisonally _append_room_entry_to_chunk - # won't append to the chunk and so we need to loop again. - # We don't want to scan over the entire range either as that - # would potentially waste a lot of work. - yield concurrently_execute( - lambda r: self._append_room_entry_to_chunk( - r, rooms_to_num_joined[r], - chunk, limit, search_filter - ), - rooms_to_scan[i:i + step], 10 - ) - if len(chunk) >= limit + 1: - break else: + step = len(rooms_to_scan) + + chunk = [] + for i in xrange(0, len(rooms_to_scan), step): + batch = rooms_to_scan[i:i + step] + logger.info("Processing %i rooms for result", len(batch)) yield concurrently_execute( lambda r: self._append_room_entry_to_chunk( r, rooms_to_num_joined[r], chunk, limit, search_filter ), - rooms_to_scan, 5 + batch, 5, ) + logger.info("Now %i rooms in result", len(chunk)) + if len(chunk) >= limit + 1: + break chunk.sort(key=lambda e: (-e["num_joined_members"], e["room_id"])) diff --git a/synapse/push/httppusher.py b/synapse/push/httppusher.py index 74c0bc462c..c16f61452c 100644 --- a/synapse/push/httppusher.py +++ b/synapse/push/httppusher.py @@ -1,5 +1,6 @@ # -*- coding: utf-8 -*- # Copyright 2015, 2016 OpenMarket Ltd +# Copyright 2017 New Vector Ltd # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. @@ -295,7 +296,7 @@ class HttpPusher(object): if event.type == 'm.room.member': d['notification']['membership'] = event.content['membership'] d['notification']['user_is_target'] = event.state_key == self.user_id - if not self.hs.config.push_redact_content and 'content' in event: + if self.hs.config.push_include_content and 'content' in event: d['notification']['content'] = event.content # We no longer send aliases separately, instead, we send the human diff --git a/synapse/push/pusherpool.py b/synapse/push/pusherpool.py index 7c069b662e..34cb108dcb 100644 --- a/synapse/push/pusherpool.py +++ b/synapse/push/pusherpool.py @@ -17,7 +17,7 @@ from twisted.internet import defer from .pusher import PusherFactory -from synapse.util.logcontext import preserve_fn, preserve_context_over_deferred +from synapse.util.logcontext import make_deferred_yieldable, preserve_fn from synapse.util.async import run_on_reactor import logging @@ -136,7 +136,7 @@ class PusherPool: ) ) - yield preserve_context_over_deferred(defer.gatherResults(deferreds)) + yield make_deferred_yieldable(defer.gatherResults(deferreds)) except Exception: logger.exception("Exception in pusher on_new_notifications") @@ -161,7 +161,7 @@ class PusherPool: preserve_fn(p.on_new_receipts)(min_stream_id, max_stream_id) ) - yield preserve_context_over_deferred(defer.gatherResults(deferreds)) + yield make_deferred_yieldable(defer.gatherResults(deferreds)) except Exception: logger.exception("Exception in pusher on_new_receipts") diff --git a/synapse/replication/slave/storage/events.py b/synapse/replication/slave/storage/events.py index 94ebbffc1b..29d7296b43 100644 --- a/synapse/replication/slave/storage/events.py +++ b/synapse/replication/slave/storage/events.py @@ -12,20 +12,18 @@ # 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 ._base import BaseSlavedStore -from ._slaved_id_tracker import SlavedIdTracker +import logging from synapse.api.constants import EventTypes from synapse.storage import DataStore -from synapse.storage.roommember import RoomMemberStore from synapse.storage.event_federation import EventFederationStore from synapse.storage.event_push_actions import EventPushActionsStore -from synapse.storage.state import StateStore +from synapse.storage.roommember import RoomMemberStore +from synapse.storage.state import StateGroupReadStore from synapse.storage.stream import StreamStore from synapse.util.caches.stream_change_cache import StreamChangeCache - -import logging - +from ._base import BaseSlavedStore +from ._slaved_id_tracker import SlavedIdTracker logger = logging.getLogger(__name__) @@ -39,7 +37,7 @@ logger = logging.getLogger(__name__) # the method descriptor on the DataStore and chuck them into our class. -class SlavedEventStore(BaseSlavedStore): +class SlavedEventStore(StateGroupReadStore, BaseSlavedStore): def __init__(self, db_conn, hs): super(SlavedEventStore, self).__init__(db_conn, hs) @@ -90,25 +88,9 @@ class SlavedEventStore(BaseSlavedStore): _get_unread_counts_by_pos_txn = ( DataStore._get_unread_counts_by_pos_txn.__func__ ) - _get_state_group_for_events = ( - StateStore.__dict__["_get_state_group_for_events"] - ) - _get_state_group_for_event = ( - StateStore.__dict__["_get_state_group_for_event"] - ) - _get_state_groups_from_groups = ( - StateStore.__dict__["_get_state_groups_from_groups"] - ) - _get_state_groups_from_groups_txn = ( - DataStore._get_state_groups_from_groups_txn.__func__ - ) get_recent_event_ids_for_room = ( StreamStore.__dict__["get_recent_event_ids_for_room"] ) - get_current_state_ids = ( - StateStore.__dict__["get_current_state_ids"] - ) - get_state_group_delta = StateStore.__dict__["get_state_group_delta"] _get_joined_hosts_cache = RoomMemberStore.__dict__["_get_joined_hosts_cache"] has_room_changed_since = DataStore.has_room_changed_since.__func__ @@ -134,12 +116,6 @@ class SlavedEventStore(BaseSlavedStore): DataStore.get_room_events_stream_for_room.__func__ ) get_events_around = DataStore.get_events_around.__func__ - get_state_for_event = DataStore.get_state_for_event.__func__ - get_state_for_events = DataStore.get_state_for_events.__func__ - get_state_groups = DataStore.get_state_groups.__func__ - get_state_groups_ids = DataStore.get_state_groups_ids.__func__ - get_state_ids_for_event = DataStore.get_state_ids_for_event.__func__ - get_state_ids_for_events = DataStore.get_state_ids_for_events.__func__ get_joined_users_from_state = DataStore.get_joined_users_from_state.__func__ get_joined_users_from_context = DataStore.get_joined_users_from_context.__func__ _get_joined_users_from_context = ( @@ -169,10 +145,7 @@ class SlavedEventStore(BaseSlavedStore): _get_rooms_for_user_where_membership_is_txn = ( DataStore._get_rooms_for_user_where_membership_is_txn.__func__ ) - _get_state_for_groups = DataStore._get_state_for_groups.__func__ - _get_all_state_from_cache = DataStore._get_all_state_from_cache.__func__ _get_events_around_txn = DataStore._get_events_around_txn.__func__ - _get_some_state_from_cache = DataStore._get_some_state_from_cache.__func__ get_backfill_events = DataStore.get_backfill_events.__func__ _get_backfill_events = DataStore._get_backfill_events.__func__ diff --git a/synapse/rest/client/versions.py b/synapse/rest/client/versions.py index e984ea47db..2ecb15deee 100644 --- a/synapse/rest/client/versions.py +++ b/synapse/rest/client/versions.py @@ -30,6 +30,7 @@ class VersionsRestServlet(RestServlet): "r0.0.1", "r0.1.0", "r0.2.0", + "r0.3.0", ] }) diff --git a/synapse/storage/_base.py b/synapse/storage/_base.py index e94917d9cd..e6eefdd6fe 100644 --- a/synapse/storage/_base.py +++ b/synapse/storage/_base.py @@ -16,8 +16,6 @@ import logging from synapse.api.errors import StoreError from synapse.util.logcontext import LoggingContext, PreserveLoggingContext -from synapse.util.caches import CACHE_SIZE_FACTOR -from synapse.util.caches.dictionary_cache import DictionaryCache from synapse.util.caches.descriptors import Cache from synapse.storage.engines import PostgresEngine import synapse.metrics @@ -180,10 +178,6 @@ class SQLBaseStore(object): self._get_event_cache = Cache("*getEvent*", keylen=3, max_entries=hs.config.event_cache_size) - self._state_group_cache = DictionaryCache( - "*stateGroupCache*", 100000 * CACHE_SIZE_FACTOR - ) - self._event_fetch_lock = threading.Condition() self._event_fetch_list = [] self._event_fetch_ongoing = 0 @@ -475,23 +469,46 @@ class SQLBaseStore(object): txn.executemany(sql, vals) + @defer.inlineCallbacks def _simple_upsert(self, table, keyvalues, values, insertion_values={}, desc="_simple_upsert", lock=True): """ + + `lock` should generally be set to True (the default), but can be set + to False if either of the following are true: + + * there is a UNIQUE INDEX on the key columns. In this case a conflict + will cause an IntegrityError in which case this function will retry + the update. + + * we somehow know that we are the only thread which will be updating + this table. + Args: table (str): The table to upsert into keyvalues (dict): The unique key tables and their new values values (dict): The nonunique columns and their new values - insertion_values (dict): key/values to use when inserting + insertion_values (dict): additional key/values to use only when + inserting + lock (bool): True to lock the table when doing the upsert. Returns: Deferred(bool): True if a new entry was created, False if an existing one was updated. """ - return self.runInteraction( - desc, - self._simple_upsert_txn, table, keyvalues, values, insertion_values, - lock - ) + while True: + try: + result = yield self.runInteraction( + desc, + self._simple_upsert_txn, table, keyvalues, values, insertion_values, + lock=lock + ) + defer.returnValue(result) + except self.database_engine.module.IntegrityError as e: + # presumably we raced with another transaction: let's retry. + logger.warn( + "IntegrityError when upserting into %s; retrying: %s", + table, e + ) def _simple_upsert_txn(self, txn, table, keyvalues, values, insertion_values={}, lock=True): @@ -499,7 +516,7 @@ class SQLBaseStore(object): if lock: self.database_engine.lock_table(txn, table) - # Try to update + # First try to update. sql = "UPDATE %s SET %s WHERE %s" % ( table, ", ".join("%s = ?" % (k,) for k in values), @@ -508,23 +525,24 @@ class SQLBaseStore(object): sqlargs = values.values() + keyvalues.values() txn.execute(sql, sqlargs) - if txn.rowcount == 0: - # We didn't update and rows so insert a new one - allvalues = {} - allvalues.update(keyvalues) - allvalues.update(values) - allvalues.update(insertion_values) + if txn.rowcount > 0: + # successfully updated at least one row. + return False - sql = "INSERT INTO %s (%s) VALUES (%s)" % ( - table, - ", ".join(k for k in allvalues), - ", ".join("?" for _ in allvalues) - ) - txn.execute(sql, allvalues.values()) + # We didn't update any rows so insert a new one + allvalues = {} + allvalues.update(keyvalues) + allvalues.update(values) + allvalues.update(insertion_values) - return True - else: - return False + sql = "INSERT INTO %s (%s) VALUES (%s)" % ( + table, + ", ".join(k for k in allvalues), + ", ".join("?" for _ in allvalues) + ) + txn.execute(sql, allvalues.values()) + # successfully inserted + return True def _simple_select_one(self, table, keyvalues, retcols, allow_none=False, desc="_simple_select_one"): diff --git a/synapse/storage/pusher.py b/synapse/storage/pusher.py index 34d2f82b7f..19ce41fde9 100644 --- a/synapse/storage/pusher.py +++ b/synapse/storage/pusher.py @@ -204,34 +204,35 @@ class PusherStore(SQLBaseStore): pushkey, pushkey_ts, lang, data, last_stream_ordering, profile_tag=""): with self._pushers_id_gen.get_next() as stream_id: - def f(txn): - newly_inserted = self._simple_upsert_txn( - txn, - "pushers", - { - "app_id": app_id, - "pushkey": pushkey, - "user_name": user_id, - }, - { - "access_token": access_token, - "kind": kind, - "app_display_name": app_display_name, - "device_display_name": device_display_name, - "ts": pushkey_ts, - "lang": lang, - "data": encode_canonical_json(data), - "last_stream_ordering": last_stream_ordering, - "profile_tag": profile_tag, - "id": stream_id, - }, - ) - if newly_inserted: - # get_if_user_has_pusher only cares if the user has - # at least *one* pusher. - txn.call_after(self.get_if_user_has_pusher.invalidate, (user_id,)) + # no need to lock because `pushers` has a unique key on + # (app_id, pushkey, user_name) so _simple_upsert will retry + newly_inserted = yield self._simple_upsert( + table="pushers", + keyvalues={ + "app_id": app_id, + "pushkey": pushkey, + "user_name": user_id, + }, + values={ + "access_token": access_token, + "kind": kind, + "app_display_name": app_display_name, + "device_display_name": device_display_name, + "ts": pushkey_ts, + "lang": lang, + "data": encode_canonical_json(data), + "last_stream_ordering": last_stream_ordering, + "profile_tag": profile_tag, + "id": stream_id, + }, + desc="add_pusher", + lock=False, + ) - yield self.runInteraction("add_pusher", f) + if newly_inserted: + # get_if_user_has_pusher only cares if the user has + # at least *one* pusher. + self.get_if_user_has_pusher.invalidate(user_id,) @defer.inlineCallbacks def delete_pusher_by_app_id_pushkey_user_id(self, app_id, pushkey, user_id): diff --git a/synapse/storage/state.py b/synapse/storage/state.py index dd01b68762..360e3e4355 100644 --- a/synapse/storage/state.py +++ b/synapse/storage/state.py @@ -13,16 +13,18 @@ # See the License for the specific language governing permissions and # limitations under the License. -from ._base import SQLBaseStore -from synapse.util.caches.descriptors import cached, cachedList -from synapse.util.caches import intern_string -from synapse.util.stringutils import to_ascii -from synapse.storage.engines import PostgresEngine +from collections import namedtuple +import logging from twisted.internet import defer -from collections import namedtuple -import logging +from synapse.storage.background_updates import BackgroundUpdateStore +from synapse.storage.engines import PostgresEngine +from synapse.util.caches import intern_string, CACHE_SIZE_FACTOR +from synapse.util.caches.descriptors import cached, cachedList +from synapse.util.caches.dictionary_cache import DictionaryCache +from synapse.util.stringutils import to_ascii +from ._base import SQLBaseStore logger = logging.getLogger(__name__) @@ -40,23 +42,11 @@ class _GetStateGroupDelta(namedtuple("_GetStateGroupDelta", ("prev_group", "delt return len(self.delta_ids) if self.delta_ids else 0 -class StateStore(SQLBaseStore): - """ Keeps track of the state at a given event. +class StateGroupReadStore(SQLBaseStore): + """The read-only parts of StateGroupStore - This is done by the concept of `state groups`. Every event is a assigned - a state group (identified by an arbitrary string), which references a - collection of state events. The current state of an event is then the - collection of state events referenced by the event's state group. - - Hence, every change in the current state causes a new state group to be - generated. However, if no change happens (e.g., if we get a message event - with only one parent it inherits the state group from its parent.) - - There are three tables: - * `state_groups`: Stores group name, first event with in the group and - room id. - * `event_to_state_groups`: Maps events to state groups. - * `state_groups_state`: Maps state group to state events. + None of these functions write to the state tables, so are suitable for + including in the SlavedStores. """ STATE_GROUP_DEDUPLICATION_UPDATE_NAME = "state_group_state_deduplication" @@ -64,21 +54,10 @@ class StateStore(SQLBaseStore): CURRENT_STATE_INDEX_UPDATE_NAME = "current_state_members_idx" def __init__(self, db_conn, hs): - super(StateStore, self).__init__(db_conn, hs) - self.register_background_update_handler( - self.STATE_GROUP_DEDUPLICATION_UPDATE_NAME, - self._background_deduplicate_state, - ) - self.register_background_update_handler( - self.STATE_GROUP_INDEX_UPDATE_NAME, - self._background_index_state, - ) - self.register_background_index_update( - self.CURRENT_STATE_INDEX_UPDATE_NAME, - index_name="current_state_events_member_index", - table="current_state_events", - columns=["state_key"], - where_clause="type='m.room.member'", + super(StateGroupReadStore, self).__init__(db_conn, hs) + + self._state_group_cache = DictionaryCache( + "*stateGroupCache*", 100000 * CACHE_SIZE_FACTOR ) @cached(max_entries=100000, iterable=True) @@ -190,178 +169,6 @@ class StateStore(SQLBaseStore): for group, event_id_map in group_to_ids.iteritems() }) - def _have_persisted_state_group_txn(self, txn, state_group): - txn.execute( - "SELECT count(*) FROM state_groups WHERE id = ?", - (state_group,) - ) - row = txn.fetchone() - return row and row[0] - - def _store_mult_state_groups_txn(self, txn, events_and_contexts): - state_groups = {} - for event, context in events_and_contexts: - if event.internal_metadata.is_outlier(): - continue - - if context.current_state_ids is None: - # AFAIK, this can never happen - logger.error( - "Non-outlier event %s had current_state_ids==None", - event.event_id) - continue - - # if the event was rejected, just give it the same state as its - # predecessor. - if context.rejected: - state_groups[event.event_id] = context.prev_group - continue - - state_groups[event.event_id] = context.state_group - - if self._have_persisted_state_group_txn(txn, context.state_group): - continue - - self._simple_insert_txn( - txn, - table="state_groups", - values={ - "id": context.state_group, - "room_id": event.room_id, - "event_id": event.event_id, - }, - ) - - # We persist as a delta if we can, while also ensuring the chain - # of deltas isn't tooo long, as otherwise read performance degrades. - if context.prev_group: - is_in_db = self._simple_select_one_onecol_txn( - txn, - table="state_groups", - keyvalues={"id": context.prev_group}, - retcol="id", - allow_none=True, - ) - if not is_in_db: - raise Exception( - "Trying to persist state with unpersisted prev_group: %r" - % (context.prev_group,) - ) - - potential_hops = self._count_state_group_hops_txn( - txn, context.prev_group - ) - if context.prev_group and potential_hops < MAX_STATE_DELTA_HOPS: - self._simple_insert_txn( - txn, - table="state_group_edges", - values={ - "state_group": context.state_group, - "prev_state_group": context.prev_group, - }, - ) - - self._simple_insert_many_txn( - txn, - table="state_groups_state", - values=[ - { - "state_group": context.state_group, - "room_id": event.room_id, - "type": key[0], - "state_key": key[1], - "event_id": state_id, - } - for key, state_id in context.delta_ids.iteritems() - ], - ) - else: - self._simple_insert_many_txn( - txn, - table="state_groups_state", - values=[ - { - "state_group": context.state_group, - "room_id": event.room_id, - "type": key[0], - "state_key": key[1], - "event_id": state_id, - } - for key, state_id in context.current_state_ids.iteritems() - ], - ) - - # Prefill the state group cache with this group. - # It's fine to use the sequence like this as the state group map - # is immutable. (If the map wasn't immutable then this prefill could - # race with another update) - txn.call_after( - self._state_group_cache.update, - self._state_group_cache.sequence, - key=context.state_group, - value=dict(context.current_state_ids), - full=True, - ) - - self._simple_insert_many_txn( - txn, - table="event_to_state_groups", - values=[ - { - "state_group": state_group_id, - "event_id": event_id, - } - for event_id, state_group_id in state_groups.iteritems() - ], - ) - - for event_id, state_group_id in state_groups.iteritems(): - txn.call_after( - self._get_state_group_for_event.prefill, - (event_id,), state_group_id - ) - - def _count_state_group_hops_txn(self, txn, state_group): - """Given a state group, count how many hops there are in the tree. - - This is used to ensure the delta chains don't get too long. - """ - if isinstance(self.database_engine, PostgresEngine): - sql = (""" - WITH RECURSIVE state(state_group) AS ( - VALUES(?::bigint) - UNION ALL - SELECT prev_state_group FROM state_group_edges e, state s - WHERE s.state_group = e.state_group - ) - SELECT count(*) FROM state; - """) - - txn.execute(sql, (state_group,)) - row = txn.fetchone() - if row and row[0]: - return row[0] - else: - return 0 - else: - # We don't use WITH RECURSIVE on sqlite3 as there are distributions - # that ship with an sqlite3 version that doesn't support it (e.g. wheezy) - next_group = state_group - count = 0 - - while next_group: - next_group = self._simple_select_one_onecol_txn( - txn, - table="state_group_edges", - keyvalues={"state_group": next_group}, - retcol="prev_state_group", - allow_none=True, - ) - if next_group: - count += 1 - - return count - @defer.inlineCallbacks def _get_state_groups_from_groups(self, groups, types): """Returns dictionary state_group -> (dict of (type, state_key) -> event id) @@ -742,6 +549,220 @@ class StateStore(SQLBaseStore): defer.returnValue(results) + +class StateStore(StateGroupReadStore, BackgroundUpdateStore): + """ Keeps track of the state at a given event. + + This is done by the concept of `state groups`. Every event is a assigned + a state group (identified by an arbitrary string), which references a + collection of state events. The current state of an event is then the + collection of state events referenced by the event's state group. + + Hence, every change in the current state causes a new state group to be + generated. However, if no change happens (e.g., if we get a message event + with only one parent it inherits the state group from its parent.) + + There are three tables: + * `state_groups`: Stores group name, first event with in the group and + room id. + * `event_to_state_groups`: Maps events to state groups. + * `state_groups_state`: Maps state group to state events. + """ + + STATE_GROUP_DEDUPLICATION_UPDATE_NAME = "state_group_state_deduplication" + STATE_GROUP_INDEX_UPDATE_NAME = "state_group_state_type_index" + CURRENT_STATE_INDEX_UPDATE_NAME = "current_state_members_idx" + + def __init__(self, db_conn, hs): + super(StateStore, self).__init__(db_conn, hs) + self.register_background_update_handler( + self.STATE_GROUP_DEDUPLICATION_UPDATE_NAME, + self._background_deduplicate_state, + ) + self.register_background_update_handler( + self.STATE_GROUP_INDEX_UPDATE_NAME, + self._background_index_state, + ) + self.register_background_index_update( + self.CURRENT_STATE_INDEX_UPDATE_NAME, + index_name="current_state_events_member_index", + table="current_state_events", + columns=["state_key"], + where_clause="type='m.room.member'", + ) + + def _have_persisted_state_group_txn(self, txn, state_group): + txn.execute( + "SELECT count(*) FROM state_groups WHERE id = ?", + (state_group,) + ) + row = txn.fetchone() + return row and row[0] + + def _store_mult_state_groups_txn(self, txn, events_and_contexts): + state_groups = {} + for event, context in events_and_contexts: + if event.internal_metadata.is_outlier(): + continue + + if context.current_state_ids is None: + # AFAIK, this can never happen + logger.error( + "Non-outlier event %s had current_state_ids==None", + event.event_id) + continue + + # if the event was rejected, just give it the same state as its + # predecessor. + if context.rejected: + state_groups[event.event_id] = context.prev_group + continue + + state_groups[event.event_id] = context.state_group + + if self._have_persisted_state_group_txn(txn, context.state_group): + continue + + self._simple_insert_txn( + txn, + table="state_groups", + values={ + "id": context.state_group, + "room_id": event.room_id, + "event_id": event.event_id, + }, + ) + + # We persist as a delta if we can, while also ensuring the chain + # of deltas isn't tooo long, as otherwise read performance degrades. + if context.prev_group: + is_in_db = self._simple_select_one_onecol_txn( + txn, + table="state_groups", + keyvalues={"id": context.prev_group}, + retcol="id", + allow_none=True, + ) + if not is_in_db: + raise Exception( + "Trying to persist state with unpersisted prev_group: %r" + % (context.prev_group,) + ) + + potential_hops = self._count_state_group_hops_txn( + txn, context.prev_group + ) + if context.prev_group and potential_hops < MAX_STATE_DELTA_HOPS: + self._simple_insert_txn( + txn, + table="state_group_edges", + values={ + "state_group": context.state_group, + "prev_state_group": context.prev_group, + }, + ) + + self._simple_insert_many_txn( + txn, + table="state_groups_state", + values=[ + { + "state_group": context.state_group, + "room_id": event.room_id, + "type": key[0], + "state_key": key[1], + "event_id": state_id, + } + for key, state_id in context.delta_ids.iteritems() + ], + ) + else: + self._simple_insert_many_txn( + txn, + table="state_groups_state", + values=[ + { + "state_group": context.state_group, + "room_id": event.room_id, + "type": key[0], + "state_key": key[1], + "event_id": state_id, + } + for key, state_id in context.current_state_ids.iteritems() + ], + ) + + # Prefill the state group cache with this group. + # It's fine to use the sequence like this as the state group map + # is immutable. (If the map wasn't immutable then this prefill could + # race with another update) + txn.call_after( + self._state_group_cache.update, + self._state_group_cache.sequence, + key=context.state_group, + value=dict(context.current_state_ids), + full=True, + ) + + self._simple_insert_many_txn( + txn, + table="event_to_state_groups", + values=[ + { + "state_group": state_group_id, + "event_id": event_id, + } + for event_id, state_group_id in state_groups.iteritems() + ], + ) + + for event_id, state_group_id in state_groups.iteritems(): + txn.call_after( + self._get_state_group_for_event.prefill, + (event_id,), state_group_id + ) + + def _count_state_group_hops_txn(self, txn, state_group): + """Given a state group, count how many hops there are in the tree. + + This is used to ensure the delta chains don't get too long. + """ + if isinstance(self.database_engine, PostgresEngine): + sql = (""" + WITH RECURSIVE state(state_group) AS ( + VALUES(?::bigint) + UNION ALL + SELECT prev_state_group FROM state_group_edges e, state s + WHERE s.state_group = e.state_group + ) + SELECT count(*) FROM state; + """) + + txn.execute(sql, (state_group,)) + row = txn.fetchone() + if row and row[0]: + return row[0] + else: + return 0 + else: + # We don't use WITH RECURSIVE on sqlite3 as there are distributions + # that ship with an sqlite3 version that doesn't support it (e.g. wheezy) + next_group = state_group + count = 0 + + while next_group: + next_group = self._simple_select_one_onecol_txn( + txn, + table="state_group_edges", + keyvalues={"state_group": next_group}, + retcol="prev_state_group", + allow_none=True, + ) + if next_group: + count += 1 + + return count + def get_next_state_group(self): return self._state_groups_id_gen.get_next() diff --git a/synapse/storage/stream.py b/synapse/storage/stream.py index dddd5fc0e7..52bdce5be2 100644 --- a/synapse/storage/stream.py +++ b/synapse/storage/stream.py @@ -39,7 +39,7 @@ from ._base import SQLBaseStore from synapse.util.caches.descriptors import cached from synapse.api.constants import EventTypes from synapse.types import RoomStreamToken -from synapse.util.logcontext import preserve_fn, preserve_context_over_deferred +from synapse.util.logcontext import make_deferred_yieldable, preserve_fn from synapse.storage.engines import PostgresEngine, Sqlite3Engine import logging @@ -234,7 +234,7 @@ class StreamStore(SQLBaseStore): results = {} room_ids = list(room_ids) for rm_ids in (room_ids[i:i + 20] for i in xrange(0, len(room_ids), 20)): - res = yield preserve_context_over_deferred(defer.gatherResults([ + res = yield make_deferred_yieldable(defer.gatherResults([ preserve_fn(self.get_room_events_stream_for_room)( room_id, from_key, to_key, limit, order=order, ) diff --git a/synapse/util/async.py b/synapse/util/async.py index e786fb38a9..0729bb2863 100644 --- a/synapse/util/async.py +++ b/synapse/util/async.py @@ -17,7 +17,7 @@ from twisted.internet import defer, reactor from .logcontext import ( - PreserveLoggingContext, preserve_fn, preserve_context_over_deferred, + PreserveLoggingContext, make_deferred_yieldable, preserve_fn ) from synapse.util import logcontext, unwrapFirstError @@ -351,7 +351,7 @@ class ReadWriteLock(object): # We wait for the latest writer to finish writing. We can safely ignore # any existing readers... as they're readers. - yield curr_writer + yield make_deferred_yieldable(curr_writer) @contextmanager def _ctx_manager(): @@ -380,7 +380,7 @@ class ReadWriteLock(object): curr_readers.clear() self.key_to_current_writer[key] = new_defer - yield preserve_context_over_deferred(defer.gatherResults(to_wait_on)) + yield make_deferred_yieldable(defer.gatherResults(to_wait_on)) @contextmanager def _ctx_manager(): diff --git a/synapse/util/distributor.py b/synapse/util/distributor.py index e68f94ce77..734331caaa 100644 --- a/synapse/util/distributor.py +++ b/synapse/util/distributor.py @@ -13,32 +13,24 @@ # See the License for the specific language governing permissions and # limitations under the License. -from twisted.internet import defer +import logging -from synapse.util.logcontext import ( - PreserveLoggingContext, preserve_context_over_fn -) +from twisted.internet import defer from synapse.util import unwrapFirstError - -import logging - +from synapse.util.logcontext import PreserveLoggingContext logger = logging.getLogger(__name__) def user_left_room(distributor, user, room_id): - return preserve_context_over_fn( - distributor.fire, - "user_left_room", user=user, room_id=room_id - ) + with PreserveLoggingContext(): + distributor.fire("user_left_room", user=user, room_id=room_id) def user_joined_room(distributor, user, room_id): - return preserve_context_over_fn( - distributor.fire, - "user_joined_room", user=user, room_id=room_id - ) + with PreserveLoggingContext(): + distributor.fire("user_joined_room", user=user, room_id=room_id) class Distributor(object): diff --git a/synapse/util/logcontext.py b/synapse/util/logcontext.py index 9683cc7265..48c9f6802d 100644 --- a/synapse/util/logcontext.py +++ b/synapse/util/logcontext.py @@ -261,67 +261,6 @@ class PreserveLoggingContext(object): ) -class _PreservingContextDeferred(defer.Deferred): - """A deferred that ensures that all callbacks and errbacks are called with - the given logging context. - """ - def __init__(self, context): - self._log_context = context - defer.Deferred.__init__(self) - - def addCallbacks(self, callback, errback=None, - callbackArgs=None, callbackKeywords=None, - errbackArgs=None, errbackKeywords=None): - callback = self._wrap_callback(callback) - errback = self._wrap_callback(errback) - return defer.Deferred.addCallbacks( - self, callback, - errback=errback, - callbackArgs=callbackArgs, - callbackKeywords=callbackKeywords, - errbackArgs=errbackArgs, - errbackKeywords=errbackKeywords, - ) - - def _wrap_callback(self, f): - def g(res, *args, **kwargs): - with PreserveLoggingContext(self._log_context): - res = f(res, *args, **kwargs) - return res - return g - - -def preserve_context_over_fn(fn, *args, **kwargs): - """Takes a function and invokes it with the given arguments, but removes - and restores the current logging context while doing so. - - If the result is a deferred, call preserve_context_over_deferred before - returning it. - """ - with PreserveLoggingContext(): - res = fn(*args, **kwargs) - - if isinstance(res, defer.Deferred): - return preserve_context_over_deferred(res) - else: - return res - - -def preserve_context_over_deferred(deferred, context=None): - """Given a deferred wrap it such that any callbacks added later to it will - be invoked with the current context. - - Deprecated: this almost certainly doesn't do want you want, ie make - the deferred follow the synapse logcontext rules: try - ``make_deferred_yieldable`` instead. - """ - if context is None: - context = LoggingContext.current_context() - d = _PreservingContextDeferred(context) - deferred.chainDeferred(d) - return d - - def preserve_fn(f): """Wraps a function, to ensure that the current context is restored after return from the function, and that the sentinel context is set once the diff --git a/synapse/visibility.py b/synapse/visibility.py index d7dbdc77ff..aaca2c584c 100644 --- a/synapse/visibility.py +++ b/synapse/visibility.py @@ -17,7 +17,7 @@ from twisted.internet import defer from synapse.api.constants import Membership, EventTypes -from synapse.util.logcontext import preserve_fn, preserve_context_over_deferred +from synapse.util.logcontext import make_deferred_yieldable, preserve_fn import logging @@ -58,7 +58,7 @@ def filter_events_for_clients(store, user_tuples, events, event_id_to_state, always_include_ids (set(event_id)): set of event ids to specifically include (unless sender is ignored) """ - forgotten = yield preserve_context_over_deferred(defer.gatherResults([ + forgotten = yield make_deferred_yieldable(defer.gatherResults([ defer.maybeDeferred( preserve_fn(store.who_forgot_in_room), room_id, |