diff options
Diffstat (limited to 'synapse')
38 files changed, 772 insertions, 359 deletions
diff --git a/synapse/__init__.py b/synapse/__init__.py index 8587ffa76f..1c27d68009 100644 --- a/synapse/__init__.py +++ b/synapse/__init__.py @@ -14,7 +14,7 @@ # See the License for the specific language governing permissions and # limitations under the License. -""" This is a reference implementation of a Matrix home server. +""" This is a reference implementation of a Matrix homeserver. """ import os @@ -36,7 +36,7 @@ try: except ImportError: pass -__version__ = "1.5.0" +__version__ = "1.5.1" if bool(os.environ.get("SYNAPSE_TEST_PATCH_LOG_CONTEXTS", False)): # We import here so that we don't have to install a bunch of deps when diff --git a/synapse/_scripts/register_new_matrix_user.py b/synapse/_scripts/register_new_matrix_user.py index bdcd915bbe..d528450c78 100644 --- a/synapse/_scripts/register_new_matrix_user.py +++ b/synapse/_scripts/register_new_matrix_user.py @@ -144,8 +144,8 @@ def main(): logging.captureWarnings(True) parser = argparse.ArgumentParser( - description="Used to register new users with a given home server when" - " registration has been disabled. The home server must be" + description="Used to register new users with a given homeserver when" + " registration has been disabled. The homeserver must be" " configured with the 'registration_shared_secret' option" " set." ) @@ -202,7 +202,7 @@ def main(): "server_url", default="https://localhost:8448", nargs="?", - help="URL to use to talk to the home server. Defaults to " + help="URL to use to talk to the homeserver. Defaults to " " 'https://localhost:8448'.", ) diff --git a/synapse/api/errors.py b/synapse/api/errors.py index cca92c34ba..5853a54c95 100644 --- a/synapse/api/errors.py +++ b/synapse/api/errors.py @@ -457,7 +457,7 @@ def cs_error(msg, code=Codes.UNKNOWN, **kwargs): class FederationError(RuntimeError): - """ This class is used to inform remote home servers about erroneous + """ This class is used to inform remote homeservers about erroneous PDUs they sent us. FATAL: The remote server could not interpret the source event. diff --git a/synapse/app/homeserver.py b/synapse/app/homeserver.py index 00a7f8330e..73e2c29d06 100644 --- a/synapse/app/homeserver.py +++ b/synapse/app/homeserver.py @@ -636,7 +636,7 @@ def run(hs): if hs.config.report_stats: logger.info("Scheduling stats reporting for 3 hour intervals") - clock.looping_call(start_phone_stats_home, 3 * 60 * 60 * 1000, hs, stats) + clock.looping_call(start_phone_stats_home, 3 * 60 * 60 * 1000) # We need to defer this init for the cases that we daemonize # otherwise the process ID we get is that of the non-daemon process @@ -644,7 +644,7 @@ def run(hs): # We wait 5 minutes to send the first set of stats as the server can # be quite busy the first few minutes - clock.call_later(5 * 60, start_phone_stats_home, hs, stats) + clock.call_later(5 * 60, start_phone_stats_home) _base.start_reactor( "synapse-homeserver", diff --git a/synapse/config/captcha.py b/synapse/config/captcha.py index 44bd5c6799..f0171bb5b2 100644 --- a/synapse/config/captcha.py +++ b/synapse/config/captcha.py @@ -35,11 +35,11 @@ class CaptchaConfig(Config): ## Captcha ## # See docs/CAPTCHA_SETUP for full details of configuring this. - # This Home Server's ReCAPTCHA public key. + # This homeserver's ReCAPTCHA public key. # #recaptcha_public_key: "YOUR_PUBLIC_KEY" - # This Home Server's ReCAPTCHA private key. + # This homeserver's ReCAPTCHA private key. # #recaptcha_private_key: "YOUR_PRIVATE_KEY" diff --git a/synapse/config/emailconfig.py b/synapse/config/emailconfig.py index 39e7a1dddb..43fad0bf8b 100644 --- a/synapse/config/emailconfig.py +++ b/synapse/config/emailconfig.py @@ -305,7 +305,7 @@ class EmailConfig(Config): # smtp_user: "exampleusername" # smtp_pass: "examplepassword" # require_transport_security: false - # notif_from: "Your Friendly %(app)s Home Server <noreply@example.com>" + # notif_from: "Your Friendly %(app)s homeserver <noreply@example.com>" # app_name: Matrix # # # Enable email notifications by default diff --git a/synapse/config/server.py b/synapse/config/server.py index d556df308d..00d01c43af 100644 --- a/synapse/config/server.py +++ b/synapse/config/server.py @@ -41,7 +41,7 @@ logger = logging.Logger(__name__) # in the list. DEFAULT_BIND_ADDRESSES = ["::", "0.0.0.0"] -DEFAULT_ROOM_VERSION = "4" +DEFAULT_ROOM_VERSION = "5" ROOM_COMPLEXITY_TOO_GREAT = ( "Your homeserver is unable to join rooms this large or complex. " @@ -721,7 +721,7 @@ class ServerConfig(Config): # Used by phonehome stats to group together related servers. #server_context: context - # Resource-constrained Homeserver Settings + # Resource-constrained homeserver Settings # # If limit_remote_rooms.enabled is True, the room complexity will be # checked before a user joins a new remote room. If it is above @@ -781,7 +781,7 @@ class ServerConfig(Config): "--daemonize", action="store_true", default=None, - help="Daemonize the home server", + help="Daemonize the homeserver", ) server_group.add_argument( "--print-pidfile", diff --git a/synapse/events/snapshot.py b/synapse/events/snapshot.py index 5f07f6fe4b..64e898f40c 100644 --- a/synapse/events/snapshot.py +++ b/synapse/events/snapshot.py @@ -33,7 +33,7 @@ class EventContext: Attributes: rejected: A rejection reason if the event was rejected, else False - state_group: The ID of the state group for this event. Note that state events + _state_group: The ID of the state group for this event. Note that state events are persisted with a state group which includes the new event, so this is effectively the state *after* the event in question. @@ -45,10 +45,24 @@ class EventContext: For an outlier, where we don't have the state at the event, this will be None. + Note that this is a private attribute: it should be accessed via + the ``state_group`` property. + + state_group_before_event: The ID of the state group representing the state + of the room before this event. + + If this is a non-state event, this will be the same as ``state_group``. If + it's a state event, it will be the same as ``prev_group``. + + If ``state_group`` is None (ie, the event is an outlier), + ``state_group_before_event`` will always also be ``None``. + prev_group: If it is known, ``state_group``'s prev_group. Note that this being None does not necessarily mean that ``state_group`` does not have a prev_group! + If the event is a state event, this is normally the same as ``prev_group``. + If ``state_group`` is None (ie, the event is an outlier), ``prev_group`` will always also be ``None``. @@ -74,7 +88,8 @@ class EventContext: ``get_current_state_ids``. _AsyncEventContext impl calculates this on-demand: it will be None until that happens. - _prev_state_ids: The room state map, excluding this event. For a non-state + _prev_state_ids: The room state map, excluding this event - ie, the state + in ``state_group_before_event``. For a non-state event, this will be the same as _current_state_events. Note that it is a completely different thing to prev_group! @@ -88,7 +103,8 @@ class EventContext: """ rejected = attr.ib(default=False, type=Union[bool, str]) - state_group = attr.ib(default=None, type=Optional[int]) + _state_group = attr.ib(default=None, type=Optional[int]) + state_group_before_event = attr.ib(default=None, type=Optional[int]) prev_group = attr.ib(default=None, type=Optional[int]) delta_ids = attr.ib(default=None, type=Optional[Dict[Tuple[str, str], str]]) app_service = attr.ib(default=None, type=Optional[ApplicationService]) @@ -100,12 +116,18 @@ class EventContext: @staticmethod def with_state( - state_group, current_state_ids, prev_state_ids, prev_group=None, delta_ids=None + state_group, + state_group_before_event, + current_state_ids, + prev_state_ids, + prev_group=None, + delta_ids=None, ): return EventContext( current_state_ids=current_state_ids, prev_state_ids=prev_state_ids, state_group=state_group, + state_group_before_event=state_group_before_event, prev_group=prev_group, delta_ids=delta_ids, ) @@ -136,7 +158,8 @@ class EventContext: "prev_state_id": prev_state_id, "event_type": event.type, "event_state_key": event.state_key if event.is_state() else None, - "state_group": self.state_group, + "state_group": self._state_group, + "state_group_before_event": self.state_group_before_event, "rejected": self.rejected, "prev_group": self.prev_group, "delta_ids": _encode_state_dict(self.delta_ids), @@ -162,6 +185,7 @@ class EventContext: event_type=input["event_type"], event_state_key=input["event_state_key"], state_group=input["state_group"], + state_group_before_event=input["state_group_before_event"], prev_group=input["prev_group"], delta_ids=_decode_state_dict(input["delta_ids"]), rejected=input["rejected"], @@ -173,22 +197,52 @@ class EventContext: return context + @property + def state_group(self) -> Optional[int]: + """The ID of the state group for this event. + + Note that state events are persisted with a state group which includes the new + event, so this is effectively the state *after* the event in question. + + For an outlier, where we don't have the state at the event, this will be None. + + It is an error to access this for a rejected event, since rejected state should + not make it into the room state. Accessing this property will raise an exception + if ``rejected`` is set. + """ + if self.rejected: + raise RuntimeError("Attempt to access state_group of rejected event") + + return self._state_group + @defer.inlineCallbacks def get_current_state_ids(self, store): - """Gets the current state IDs + """ + Gets the room state map, including this event - ie, the state in ``state_group`` + + It is an error to access this for a rejected event, since rejected state should + not make it into the room state. This method will raise an exception if + ``rejected`` is set. Returns: Deferred[dict[(str, str), str]|None]: Returns None if state_group is None, which happens when the associated event is an outlier. + Maps a (type, state_key) to the event ID of the state event matching this tuple. """ + if self.rejected: + raise RuntimeError("Attempt to access state_ids of rejected event") + yield self._ensure_fetched(store) return self._current_state_ids @defer.inlineCallbacks def get_prev_state_ids(self, store): - """Gets the prev state IDs + """ + Gets the room state map, excluding this event. + + For a non-state event, this will be the same as get_current_state_ids(). Returns: Deferred[dict[(str, str), str]|None]: Returns None if state_group @@ -202,11 +256,17 @@ class EventContext: def get_cached_current_state_ids(self): """Gets the current state IDs if we have them already cached. + It is an error to access this for a rejected event, since rejected state should + not make it into the room state. This method will raise an exception if + ``rejected`` is set. + Returns: dict[(str, str), str]|None: Returns None if we haven't cached the state or if state_group is None, which happens when the associated event is an outlier. """ + if self.rejected: + raise RuntimeError("Attempt to access state_ids of rejected event") return self._current_state_ids diff --git a/synapse/federation/federation_client.py b/synapse/federation/federation_client.py index 545d719652..27f6aff004 100644 --- a/synapse/federation/federation_client.py +++ b/synapse/federation/federation_client.py @@ -177,7 +177,7 @@ class FederationClient(FederationBase): given destination server. Args: - dest (str): The remote home server to ask. + dest (str): The remote homeserver to ask. room_id (str): The room_id to backfill. limit (int): The maximum number of PDUs to return. extremities (list): List of PDU id and origins of the first pdus @@ -227,7 +227,7 @@ class FederationClient(FederationBase): one succeeds. Args: - destinations (list): Which home servers to query + destinations (list): Which homeservers to query event_id (str): event to fetch room_version (str): version of the room outlier (bool): Indicates whether the PDU is an `outlier`, i.e. if @@ -312,7 +312,7 @@ class FederationClient(FederationBase): @defer.inlineCallbacks @log_function def get_state_for_room(self, destination, room_id, event_id): - """Requests all of the room state at a given event from a remote home server. + """Requests all of the room state at a given event from a remote homeserver. Args: destination (str): The remote homeserver to query for the state. diff --git a/synapse/federation/transport/__init__.py b/synapse/federation/transport/__init__.py index d9fcc520a0..5db733af98 100644 --- a/synapse/federation/transport/__init__.py +++ b/synapse/federation/transport/__init__.py @@ -14,9 +14,9 @@ # limitations under the License. """The transport layer is responsible for both sending transactions to remote -home servers and receiving a variety of requests from other home servers. +homeservers and receiving a variety of requests from other homeservers. -By default this is done over HTTPS (and all home servers are required to +By default this is done over HTTPS (and all homeservers are required to support HTTPS), however individual pairings of servers may decide to communicate over a different (albeit still reliable) protocol. """ diff --git a/synapse/federation/transport/client.py b/synapse/federation/transport/client.py index 920fa86853..dc95ab2113 100644 --- a/synapse/federation/transport/client.py +++ b/synapse/federation/transport/client.py @@ -44,7 +44,7 @@ class TransportLayerClient(object): given event. Args: - destination (str): The host name of the remote home server we want + destination (str): The host name of the remote homeserver we want to get the state from. context (str): The name of the context we want the state of event_id (str): The event we want the context at. @@ -68,7 +68,7 @@ class TransportLayerClient(object): given event. Returns the state's event_id's Args: - destination (str): The host name of the remote home server we want + destination (str): The host name of the remote homeserver we want to get the state from. context (str): The name of the context we want the state of event_id (str): The event we want the context at. @@ -91,7 +91,7 @@ class TransportLayerClient(object): """ Requests the pdu with give id and origin from the given server. Args: - destination (str): The host name of the remote home server we want + destination (str): The host name of the remote homeserver we want to get the state from. event_id (str): The id of the event being requested. timeout (int): How long to try (in ms) the destination for before diff --git a/synapse/federation/transport/server.py b/synapse/federation/transport/server.py index d6c23f22bd..09baa9c57d 100644 --- a/synapse/federation/transport/server.py +++ b/synapse/federation/transport/server.py @@ -714,7 +714,7 @@ class PublicRoomList(BaseFederationServlet): This API returns information in the same format as /publicRooms on the client API, but will only ever include local public rooms and hence is - intended for consumption by other home servers. + intended for consumption by other homeservers. GET /publicRooms HTTP/1.1 diff --git a/synapse/handlers/auth.py b/synapse/handlers/auth.py index 0955cf9dba..54a71c49d2 100644 --- a/synapse/handlers/auth.py +++ b/synapse/handlers/auth.py @@ -250,7 +250,7 @@ class AuthHandler(BaseHandler): # could continue registration from your phone having clicked the # email auth link on there). It's probably too open to abuse # because it lets unauthenticated clients store arbitrary objects - # on a home server. + # on a homeserver. # Revisit: Assumimg the REST APIs do sensible validation, the data # isn't arbintrary. session["clientdict"] = clientdict @@ -817,7 +817,7 @@ class AuthHandler(BaseHandler): @defer.inlineCallbacks def add_threepid(self, user_id, medium, address, validated_at): # 'Canonicalise' email addresses down to lower case. - # We've now moving towards the Home Server being the entity that + # We've now moving towards the homeserver being the entity that # is responsible for validating threepids used for resetting passwords # on accounts, so in future Synapse will gain knowledge of specific # types (mediums) of threepid. For now, we still use the existing diff --git a/synapse/handlers/directory.py b/synapse/handlers/directory.py index c4632f8984..69051101a6 100644 --- a/synapse/handlers/directory.py +++ b/synapse/handlers/directory.py @@ -283,7 +283,7 @@ class DirectoryHandler(BaseHandler): def on_directory_query(self, args): room_alias = RoomAlias.from_string(args["room_alias"]) if not self.hs.is_mine(room_alias): - raise SynapseError(400, "Room Alias is not hosted on this Home Server") + raise SynapseError(400, "Room Alias is not hosted on this homeserver") result = yield self.get_association_from_room_alias(room_alias) diff --git a/synapse/handlers/federation.py b/synapse/handlers/federation.py index 8cafcfdab0..0e904f2da0 100644 --- a/synapse/handlers/federation.py +++ b/synapse/handlers/federation.py @@ -97,9 +97,9 @@ class FederationHandler(BaseHandler): """Handles events that originated from federation. Responsible for: a) handling received Pdus before handing them on as Events to the rest - of the home server (including auth and state conflict resoultion) + of the homeserver (including auth and state conflict resoultion) b) converting events that were produced by local clients that may need - to be sent to remote home servers. + to be sent to remote homeservers. c) doing the necessary dances to invite remote users and join remote rooms. """ @@ -1688,7 +1688,11 @@ class FederationHandler(BaseHandler): # hack around with a try/finally instead. success = False try: - if not event.internal_metadata.is_outlier() and not backfilled: + if ( + not event.internal_metadata.is_outlier() + and not backfilled + and not context.rejected + ): yield self.action_generator.handle_push_actions_for_event( event, context ) @@ -2276,6 +2280,7 @@ class FederationHandler(BaseHandler): return EventContext.with_state( state_group=state_group, + state_group_before_event=context.state_group_before_event, current_state_ids=current_state_ids, prev_state_ids=prev_state_ids, prev_group=prev_group, diff --git a/synapse/handlers/pagination.py b/synapse/handlers/pagination.py index 97f15a1c32..260a4351ca 100644 --- a/synapse/handlers/pagination.py +++ b/synapse/handlers/pagination.py @@ -127,7 +127,9 @@ class PaginationHandler(object): self._purges_in_progress_by_room.add(room_id) try: with (yield self.pagination_lock.write(room_id)): - yield self.store.purge_history(room_id, token, delete_local_events) + yield self.storage.purge_events.purge_history( + room_id, token, delete_local_events + ) logger.info("[purge] complete") self._purges_by_id[purge_id].status = PurgeStatus.STATUS_COMPLETE except Exception: @@ -170,7 +172,7 @@ class PaginationHandler(object): if joined: raise SynapseError(400, "Users are still joined to this room") - await self.store.purge_room(room_id) + await self.storage.purge_events.purge_room(room_id) @defer.inlineCallbacks def get_messages( diff --git a/synapse/handlers/profile.py b/synapse/handlers/profile.py index 22e0a04da4..1e5a4613c9 100644 --- a/synapse/handlers/profile.py +++ b/synapse/handlers/profile.py @@ -152,7 +152,7 @@ class BaseProfileHandler(BaseHandler): by_admin (bool): Whether this change was made by an administrator. """ if not self.hs.is_mine(target_user): - raise SynapseError(400, "User is not hosted on this Home Server") + raise SynapseError(400, "User is not hosted on this homeserver") if not by_admin and target_user != requester.user: raise AuthError(400, "Cannot set another user's displayname") @@ -207,7 +207,7 @@ class BaseProfileHandler(BaseHandler): """target_user is the user whose avatar_url is to be changed; auth_user is the user attempting to make this change.""" if not self.hs.is_mine(target_user): - raise SynapseError(400, "User is not hosted on this Home Server") + raise SynapseError(400, "User is not hosted on this homeserver") if not by_admin and target_user != requester.user: raise AuthError(400, "Cannot set another user's avatar_url") @@ -231,7 +231,7 @@ class BaseProfileHandler(BaseHandler): def on_profile_query(self, args): user = UserID.from_string(args["user_id"]) if not self.hs.is_mine(user): - raise SynapseError(400, "User is not hosted on this Home Server") + raise SynapseError(400, "User is not hosted on this homeserver") just_field = args.get("field", None) diff --git a/synapse/handlers/register.py b/synapse/handlers/register.py index cff6b0d375..95806af41e 100644 --- a/synapse/handlers/register.py +++ b/synapse/handlers/register.py @@ -24,7 +24,6 @@ from synapse.api.errors import ( AuthError, Codes, ConsentNotGivenError, - LimitExceededError, RegistrationError, SynapseError, ) @@ -168,6 +167,7 @@ class RegistrationHandler(BaseHandler): Raises: RegistrationError if there was a problem registering. """ + yield self.check_registration_ratelimit(address) yield self.auth.check_auth_blocking(threepid=threepid) password_hash = None @@ -217,8 +217,13 @@ class RegistrationHandler(BaseHandler): else: # autogen a sequential user ID + fail_count = 0 user = None while not user: + # Fail after being unable to find a suitable ID a few times + if fail_count > 10: + raise SynapseError(500, "Unable to find a suitable guest user ID") + localpart = yield self._generate_user_id() user = UserID(localpart, self.hs.hostname) user_id = user.to_string() @@ -233,10 +238,14 @@ class RegistrationHandler(BaseHandler): create_profile_with_displayname=default_display_name, address=address, ) + + # Successfully registered + break except SynapseError: # if user id is taken, just generate another user = None user_id = None + fail_count += 1 if not self.hs.config.user_consent_at_registration: yield self._auto_join_rooms(user_id) @@ -414,6 +423,29 @@ class RegistrationHandler(BaseHandler): ratelimit=False, ) + def check_registration_ratelimit(self, address): + """A simple helper method to check whether the registration rate limit has been hit + for a given IP address + + Args: + address (str|None): the IP address used to perform the registration. If this is + None, no ratelimiting will be performed. + + Raises: + LimitExceededError: If the rate limit has been exceeded. + """ + if not address: + return + + time_now = self.clock.time() + + self.ratelimiter.ratelimit( + address, + time_now_s=time_now, + rate_hz=self.hs.config.rc_registration.per_second, + burst_count=self.hs.config.rc_registration.burst_count, + ) + def register_with_store( self, user_id, @@ -446,22 +478,6 @@ class RegistrationHandler(BaseHandler): Returns: Deferred """ - # Don't rate limit for app services - if appservice_id is None and address is not None: - time_now = self.clock.time() - - allowed, time_allowed = self.ratelimiter.can_do_action( - address, - time_now_s=time_now, - rate_hz=self.hs.config.rc_registration.per_second, - burst_count=self.hs.config.rc_registration.burst_count, - ) - - if not allowed: - raise LimitExceededError( - retry_after_ms=int(1000 * (time_allowed - time_now)) - ) - if self.hs.config.worker_app: return self._register_client( user_id=user_id, @@ -614,7 +630,7 @@ class RegistrationHandler(BaseHandler): # And we add an email pusher for them by default, but only # if email notifications are enabled (so people don't start # getting mail spam where they weren't before if email - # notifs are set up on a home server) + # notifs are set up on a homeserver) if ( self.hs.config.email_enable_notifs and self.hs.config.email_notif_for_new_users diff --git a/synapse/handlers/room_member.py b/synapse/handlers/room_member.py index 06d09c2947..6cfee4b361 100644 --- a/synapse/handlers/room_member.py +++ b/synapse/handlers/room_member.py @@ -515,6 +515,15 @@ class RoomMemberHandler(object): yield self.store.set_room_is_public(old_room_id, False) yield self.store.set_room_is_public(room_id, True) + # Check if any groups we own contain the predecessor room + local_group_ids = yield self.store.get_local_groups_for_room(old_room_id) + for group_id in local_group_ids: + # Add new the new room to those groups + yield self.store.add_room_to_group(group_id, room_id, old_room["is_public"]) + + # Remove the old room from those groups + yield self.store.remove_room_from_group(group_id, old_room_id) + @defer.inlineCallbacks def copy_user_state_on_room_upgrade(self, old_room_id, new_room_id, user_ids): """Copy user-specific information when they join a new room when that new room is the diff --git a/synapse/handlers/typing.py b/synapse/handlers/typing.py index ca8ae9fb5b..856337b7e2 100644 --- a/synapse/handlers/typing.py +++ b/synapse/handlers/typing.py @@ -120,7 +120,7 @@ class TypingHandler(object): auth_user_id = auth_user.to_string() if not self.is_mine_id(target_user_id): - raise SynapseError(400, "User is not hosted on this Home Server") + raise SynapseError(400, "User is not hosted on this homeserver") if target_user_id != auth_user_id: raise AuthError(400, "Cannot set another user's typing state") @@ -150,7 +150,7 @@ class TypingHandler(object): auth_user_id = auth_user.to_string() if not self.is_mine_id(target_user_id): - raise SynapseError(400, "User is not hosted on this Home Server") + raise SynapseError(400, "User is not hosted on this homeserver") if target_user_id != auth_user_id: raise AuthError(400, "Cannot set another user's typing state") diff --git a/synapse/http/matrixfederationclient.py b/synapse/http/matrixfederationclient.py index 691380abda..16765d54e0 100644 --- a/synapse/http/matrixfederationclient.py +++ b/synapse/http/matrixfederationclient.py @@ -530,7 +530,7 @@ class MatrixFederationHttpClient(object): """ Builds the Authorization headers for a federation request Args: - destination (bytes|None): The desination home server of the request. + destination (bytes|None): The desination homeserver of the request. May be None if the destination is an identity server, in which case destination_is must be non-None. method (bytes): The HTTP method of the request diff --git a/synapse/logging/_terse_json.py b/synapse/logging/_terse_json.py index 0ebbde06f2..76ce7d8808 100644 --- a/synapse/logging/_terse_json.py +++ b/synapse/logging/_terse_json.py @@ -153,7 +153,7 @@ class TerseJSONToTCPLogObserver(object): An IObserver that writes JSON logs to a TCP target. Args: - hs (HomeServer): The Homeserver that is being logged for. + hs (HomeServer): The homeserver that is being logged for. host: The host of the logging target. port: The logging target's port. metadata: Metadata to be added to each log entry. diff --git a/synapse/replication/http/register.py b/synapse/replication/http/register.py index 915cfb9430..0c4aca1291 100644 --- a/synapse/replication/http/register.py +++ b/synapse/replication/http/register.py @@ -75,6 +75,8 @@ class ReplicationRegisterServlet(ReplicationEndpoint): async def _handle_request(self, request, user_id): content = parse_json_object_from_request(request) + self.registration_handler.check_registration_ratelimit(content["address"]) + await self.registration_handler.register_with_store( user_id=user_id, password_hash=content["password_hash"], diff --git a/synapse/rest/media/v1/preview_url_resource.py b/synapse/rest/media/v1/preview_url_resource.py index 531d923f76..15c15a12f5 100644 --- a/synapse/rest/media/v1/preview_url_resource.py +++ b/synapse/rest/media/v1/preview_url_resource.py @@ -56,6 +56,9 @@ logger = logging.getLogger(__name__) _charset_match = re.compile(br"<\s*meta[^>]*charset\s*=\s*([a-z0-9-]+)", flags=re.I) _content_type_match = re.compile(r'.*; *charset="?(.*?)"?(;|$)', flags=re.I) +OG_TAG_NAME_MAXLEN = 50 +OG_TAG_VALUE_MAXLEN = 1000 + class PreviewUrlResource(DirectServeResource): isLeaf = True @@ -171,7 +174,7 @@ class PreviewUrlResource(DirectServeResource): ts (int): Returns: - Deferred[str]: json-encoded og data + Deferred[bytes]: json-encoded og data """ # check the URL cache in the DB (which will also provide us with # historical previews, if we have any) @@ -272,6 +275,18 @@ class PreviewUrlResource(DirectServeResource): logger.warning("Failed to find any OG data in %s", url) og = {} + # filter out any stupidly long values + keys_to_remove = [] + for k, v in og.items(): + # values can be numeric as well as strings, hence the cast to str + if len(k) > OG_TAG_NAME_MAXLEN or len(str(v)) > OG_TAG_VALUE_MAXLEN: + logger.warning( + "Pruning overlong tag %s from OG data", k[:OG_TAG_NAME_MAXLEN] + ) + keys_to_remove.append(k) + for k in keys_to_remove: + del og[k] + logger.debug("Calculated OG for %s as %s", url, og) jsonog = json.dumps(og) @@ -506,6 +521,10 @@ def _calc_og(tree, media_uri): og = {} for tag in tree.xpath("//*/meta[starts-with(@property, 'og:')]"): if "content" in tag.attrib: + # if we've got more than 50 tags, someone is taking the piss + if len(og) >= 50: + logger.warning("Skipping OG for page with too many 'og:' tags") + return {} og[tag.attrib["property"]] = tag.attrib["content"] # TODO: grab article: meta tags too, e.g.: diff --git a/synapse/state/__init__.py b/synapse/state/__init__.py index 2c04ab1854..139beef8ed 100644 --- a/synapse/state/__init__.py +++ b/synapse/state/__init__.py @@ -16,6 +16,7 @@ import logging from collections import namedtuple +from typing import Iterable, Optional from six import iteritems, itervalues @@ -27,6 +28,7 @@ from twisted.internet import defer from synapse.api.constants import EventTypes from synapse.api.room_versions import KNOWN_ROOM_VERSIONS, StateResolutionVersions +from synapse.events import EventBase from synapse.events.snapshot import EventContext from synapse.logging.utils import log_function from synapse.state import v1, v2 @@ -212,15 +214,17 @@ class StateHandler(object): return joined_hosts @defer.inlineCallbacks - def compute_event_context(self, event, old_state=None): + def compute_event_context( + self, event: EventBase, old_state: Optional[Iterable[EventBase]] = None + ): """Build an EventContext structure for the event. This works out what the current state should be for the event, and generates a new state group if necessary. Args: - event (synapse.events.EventBase): - old_state (dict|None): The state at the event if it can't be + event: + old_state: The state at the event if it can't be calculated from existing events. This is normally only specified when receiving an event from federation where we don't have the prev events for, e.g. when backfilling. @@ -251,113 +255,103 @@ class StateHandler(object): # group for it. context = EventContext.with_state( state_group=None, + state_group_before_event=None, current_state_ids=current_state_ids, prev_state_ids=prev_state_ids, ) return context + # + # first of all, figure out the state before the event + # + if old_state: - # We already have the state, so we don't need to calculate it. - # Let's just correctly fill out the context and create a - # new state group for it. - - prev_state_ids = {(s.type, s.state_key): s.event_id for s in old_state} - - if event.is_state(): - key = (event.type, event.state_key) - if key in prev_state_ids: - replaces = prev_state_ids[key] - if replaces != event.event_id: # Paranoia check - event.unsigned["replaces_state"] = replaces - current_state_ids = dict(prev_state_ids) - current_state_ids[key] = event.event_id - else: - current_state_ids = prev_state_ids + # if we're given the state before the event, then we use that + state_ids_before_event = { + (s.type, s.state_key): s.event_id for s in old_state + } + state_group_before_event = None + state_group_before_event_prev_group = None + deltas_to_state_group_before_event = None - state_group = yield self.state_store.store_state_group( - event.event_id, - event.room_id, - prev_group=None, - delta_ids=None, - current_state_ids=current_state_ids, - ) + else: + # otherwise, we'll need to resolve the state across the prev_events. + logger.debug("calling resolve_state_groups from compute_event_context") - context = EventContext.with_state( - state_group=state_group, - current_state_ids=current_state_ids, - prev_state_ids=prev_state_ids, + entry = yield self.resolve_state_groups_for_events( + event.room_id, event.prev_event_ids() ) - return context + state_ids_before_event = entry.state + state_group_before_event = entry.state_group + state_group_before_event_prev_group = entry.prev_group + deltas_to_state_group_before_event = entry.delta_ids - logger.debug("calling resolve_state_groups from compute_event_context") + # + # make sure that we have a state group at that point. If it's not a state event, + # that will be the state group for the new event. If it *is* a state event, + # it might get rejected (in which case we'll need to persist it with the + # previous state group) + # - entry = yield self.resolve_state_groups_for_events( - event.room_id, event.prev_event_ids() - ) + if not state_group_before_event: + state_group_before_event = yield self.state_store.store_state_group( + event.event_id, + event.room_id, + prev_group=state_group_before_event_prev_group, + delta_ids=deltas_to_state_group_before_event, + current_state_ids=state_ids_before_event, + ) - prev_state_ids = entry.state - prev_group = None - delta_ids = None + # XXX: can we update the state cache entry for the new state group? or + # could we set a flag on resolve_state_groups_for_events to tell it to + # always make a state group? + + # + # now if it's not a state event, we're done + # + + if not event.is_state(): + return EventContext.with_state( + state_group_before_event=state_group_before_event, + state_group=state_group_before_event, + current_state_ids=state_ids_before_event, + prev_state_ids=state_ids_before_event, + prev_group=state_group_before_event_prev_group, + delta_ids=deltas_to_state_group_before_event, + ) - if event.is_state(): - # If this is a state event then we need to create a new state - # group for the state after this event. + # + # otherwise, we'll need to create a new state group for after the event + # - key = (event.type, event.state_key) - if key in prev_state_ids: - replaces = prev_state_ids[key] + key = (event.type, event.state_key) + if key in state_ids_before_event: + replaces = state_ids_before_event[key] + if replaces != event.event_id: event.unsigned["replaces_state"] = replaces - current_state_ids = dict(prev_state_ids) - current_state_ids[key] = event.event_id - - if entry.state_group: - # If the state at the event has a state group assigned then - # we can use that as the prev group - prev_group = entry.state_group - delta_ids = {key: event.event_id} - elif entry.prev_group: - # If the state at the event only has a prev group, then we can - # use that as a prev group too. - prev_group = entry.prev_group - delta_ids = dict(entry.delta_ids) - delta_ids[key] = event.event_id - - state_group = yield self.state_store.store_state_group( - event.event_id, - event.room_id, - prev_group=prev_group, - delta_ids=delta_ids, - current_state_ids=current_state_ids, - ) - else: - current_state_ids = prev_state_ids - prev_group = entry.prev_group - delta_ids = entry.delta_ids - - if entry.state_group is None: - entry.state_group = yield self.state_store.store_state_group( - event.event_id, - event.room_id, - prev_group=entry.prev_group, - delta_ids=entry.delta_ids, - current_state_ids=current_state_ids, - ) - entry.state_id = entry.state_group - - state_group = entry.state_group - - context = EventContext.with_state( - state_group=state_group, - current_state_ids=current_state_ids, - prev_state_ids=prev_state_ids, - prev_group=prev_group, + state_ids_after_event = dict(state_ids_before_event) + state_ids_after_event[key] = event.event_id + delta_ids = {key: event.event_id} + + state_group_after_event = yield self.state_store.store_state_group( + event.event_id, + event.room_id, + prev_group=state_group_before_event, delta_ids=delta_ids, + current_state_ids=state_ids_after_event, ) - return context + return EventContext.with_state( + state_group=state_group_after_event, + state_group_before_event=state_group_before_event, + current_state_ids=state_ids_after_event, + prev_state_ids=state_ids_before_event, + prev_group=state_group_before_event, + delta_ids=delta_ids, + ) @measure_func() @defer.inlineCallbacks diff --git a/synapse/storage/__init__.py b/synapse/storage/__init__.py index 0a1a8cc1e5..0460fe8cc9 100644 --- a/synapse/storage/__init__.py +++ b/synapse/storage/__init__.py @@ -30,6 +30,7 @@ stored in `synapse.storage.schema`. from synapse.storage.data_stores import DataStores from synapse.storage.data_stores.main import DataStore from synapse.storage.persist_events import EventsPersistenceStorage +from synapse.storage.purge_events import PurgeEventsStorage from synapse.storage.state import StateGroupStorage __all__ = ["DataStores", "DataStore"] @@ -46,6 +47,7 @@ class Storage(object): self.main = stores.main self.persistence = EventsPersistenceStorage(hs, stores) + self.purge_events = PurgeEventsStorage(hs, stores) self.state = StateGroupStorage(hs, stores) diff --git a/synapse/storage/_base.py b/synapse/storage/_base.py index 1a2b7ebe25..ab596fa68d 100644 --- a/synapse/storage/_base.py +++ b/synapse/storage/_base.py @@ -361,14 +361,11 @@ class SQLBaseStore(object): expiration_ts, ) - self._simple_insert_txn( + self._simple_upsert_txn( txn, "account_validity", - values={ - "user_id": user_id, - "expiration_ts_ms": expiration_ts, - "email_sent": False, - }, + keyvalues={"user_id": user_id}, + values={"expiration_ts_ms": expiration_ts, "email_sent": False}, ) def start_profiling(self): diff --git a/synapse/storage/data_stores/main/deviceinbox.py b/synapse/storage/data_stores/main/deviceinbox.py index f04aad0743..96cd0fb77a 100644 --- a/synapse/storage/data_stores/main/deviceinbox.py +++ b/synapse/storage/data_stores/main/deviceinbox.py @@ -358,8 +358,21 @@ class DeviceInboxStore(DeviceInboxWorkerStore, DeviceInboxBackgroundUpdateStore) def _add_messages_to_local_device_inbox_txn( self, txn, stream_id, messages_by_user_then_device ): - sql = "UPDATE device_max_stream_id" " SET stream_id = ?" " WHERE stream_id < ?" - txn.execute(sql, (stream_id, stream_id)) + # Compatible method of performing an upsert + sql = "SELECT stream_id FROM device_max_stream_id" + + txn.execute(sql) + rows = txn.fetchone() + if rows: + db_stream_id = rows[0] + if db_stream_id < stream_id: + # Insert the new stream_id + sql = "UPDATE device_max_stream_id SET stream_id = ?" + else: + # No rows, perform an insert + sql = "INSERT INTO device_max_stream_id (stream_id) VALUES (?)" + + txn.execute(sql, (stream_id,)) local_by_user_then_device = {} for user_id, messages_by_device in messages_by_user_then_device.items(): diff --git a/synapse/storage/data_stores/main/events.py b/synapse/storage/data_stores/main/events.py index 301f8ea128..878f7568a6 100644 --- a/synapse/storage/data_stores/main/events.py +++ b/synapse/storage/data_stores/main/events.py @@ -1375,6 +1375,10 @@ class EventsStore( if True, we will delete local events as well as remote ones (instead of just marking them as outliers and deleting their state groups). + + Returns: + Deferred[set[int]]: The set of state groups that are referenced by + deleted events. """ return self.runInteraction( @@ -1511,11 +1515,10 @@ class EventsStore( [(room_id, event_id) for event_id, in new_backwards_extrems], ) - logger.info("[purge] finding redundant state groups") + logger.info("[purge] finding state groups referenced by deleted events") # Get all state groups that are referenced by events that are to be - # deleted. We then go and check if they are referenced by other events - # or state groups, and if not we delete them. + # deleted. txn.execute( """ SELECT DISTINCT state_group FROM events_to_purge @@ -1528,60 +1531,6 @@ class EventsStore( "[purge] found %i referenced state groups", len(referenced_state_groups) ) - logger.info("[purge] finding state groups that can be deleted") - - _ = self._find_unreferenced_groups_during_purge(txn, referenced_state_groups) - state_groups_to_delete, remaining_state_groups = _ - - logger.info( - "[purge] found %i state groups to delete", len(state_groups_to_delete) - ) - - logger.info( - "[purge] de-delta-ing %i remaining state groups", - len(remaining_state_groups), - ) - - # Now we turn the state groups that reference to-be-deleted state - # groups to non delta versions. - for sg in remaining_state_groups: - logger.info("[purge] de-delta-ing remaining state group %s", sg) - curr_state = self._get_state_groups_from_groups_txn(txn, [sg]) - curr_state = curr_state[sg] - - self._simple_delete_txn( - txn, table="state_groups_state", keyvalues={"state_group": sg} - ) - - self._simple_delete_txn( - txn, table="state_group_edges", keyvalues={"state_group": sg} - ) - - self._simple_insert_many_txn( - txn, - table="state_groups_state", - values=[ - { - "state_group": sg, - "room_id": room_id, - "type": key[0], - "state_key": key[1], - "event_id": state_id, - } - for key, state_id in iteritems(curr_state) - ], - ) - - logger.info("[purge] removing redundant state groups") - txn.executemany( - "DELETE FROM state_groups_state WHERE state_group = ?", - ((sg,) for sg in state_groups_to_delete), - ) - txn.executemany( - "DELETE FROM state_groups WHERE id = ?", - ((sg,) for sg in state_groups_to_delete), - ) - logger.info("[purge] removing events from event_to_state_groups") txn.execute( "DELETE FROM event_to_state_groups " @@ -1668,138 +1617,35 @@ class EventsStore( logger.info("[purge] done") - def _find_unreferenced_groups_during_purge(self, txn, state_groups): - """Used when purging history to figure out which state groups can be - deleted and which need to be de-delta'ed (due to one of its prev groups - being scheduled for deletion). - - Args: - txn - state_groups (set[int]): Set of state groups referenced by events - that are going to be deleted. - - Returns: - tuple[set[int], set[int]]: The set of state groups that can be - deleted and the set of state groups that need to be de-delta'ed - """ - # Graph of state group -> previous group - graph = {} - - # Set of events that we have found to be referenced by events - referenced_groups = set() - - # Set of state groups we've already seen - state_groups_seen = set(state_groups) - - # Set of state groups to handle next. - next_to_search = set(state_groups) - while next_to_search: - # We bound size of groups we're looking up at once, to stop the - # SQL query getting too big - if len(next_to_search) < 100: - current_search = next_to_search - next_to_search = set() - else: - current_search = set(itertools.islice(next_to_search, 100)) - next_to_search -= current_search - - # Check if state groups are referenced - sql = """ - SELECT DISTINCT state_group FROM event_to_state_groups - LEFT JOIN events_to_purge AS ep USING (event_id) - WHERE ep.event_id IS NULL AND - """ - clause, args = make_in_list_sql_clause( - txn.database_engine, "state_group", current_search - ) - txn.execute(sql + clause, list(args)) - - referenced = set(sg for sg, in txn) - referenced_groups |= referenced - - # We don't continue iterating up the state group graphs for state - # groups that are referenced. - current_search -= referenced - - rows = self._simple_select_many_txn( - txn, - table="state_group_edges", - column="prev_state_group", - iterable=current_search, - keyvalues={}, - retcols=("prev_state_group", "state_group"), - ) - - prevs = set(row["state_group"] for row in rows) - # We don't bother re-handling groups we've already seen - prevs -= state_groups_seen - next_to_search |= prevs - state_groups_seen |= prevs - - for row in rows: - # Note: Each state group can have at most one prev group - graph[row["state_group"]] = row["prev_state_group"] - - to_delete = state_groups_seen - referenced_groups - - to_dedelta = set() - for sg in referenced_groups: - prev_sg = graph.get(sg) - if prev_sg and prev_sg in to_delete: - to_dedelta.add(sg) - - return to_delete, to_dedelta + return referenced_state_groups def purge_room(self, room_id): """Deletes all record of a room Args: - room_id (str): + room_id (str) + + Returns: + Deferred[List[int]]: The list of state groups to delete. """ return self.runInteraction("purge_room", self._purge_room_txn, room_id) def _purge_room_txn(self, txn, room_id): - # first we have to delete the state groups states - logger.info("[purge] removing %s from state_groups_state", room_id) - + # First we fetch all the state groups that should be deleted, before + # we delete that information. txn.execute( """ - DELETE FROM state_groups_state WHERE state_group IN ( - SELECT state_group FROM events JOIN event_to_state_groups USING(event_id) - WHERE events.room_id=? - ) + SELECT DISTINCT state_group FROM events + INNER JOIN event_to_state_groups USING(event_id) + WHERE events.room_id = ? """, (room_id,), ) - # ... and the state group edges - logger.info("[purge] removing %s from state_group_edges", room_id) - - txn.execute( - """ - DELETE FROM state_group_edges WHERE state_group IN ( - SELECT state_group FROM events JOIN event_to_state_groups USING(event_id) - WHERE events.room_id=? - ) - """, - (room_id,), - ) - - # ... and the state groups - logger.info("[purge] removing %s from state_groups", room_id) - - txn.execute( - """ - DELETE FROM state_groups WHERE id IN ( - SELECT state_group FROM events JOIN event_to_state_groups USING(event_id) - WHERE events.room_id=? - ) - """, - (room_id,), - ) + state_groups = [row[0] for row in txn] - # and then tables which lack an index on room_id but have one on event_id + # Now we delete tables which lack an index on room_id but have one on event_id for table in ( "event_auth", "event_edges", @@ -1887,6 +1733,165 @@ class EventsStore( logger.info("[purge] done") + return state_groups + + def purge_unreferenced_state_groups( + self, room_id: str, state_groups_to_delete + ) -> defer.Deferred: + """Deletes no longer referenced state groups and de-deltas any state + groups that reference them. + + Args: + room_id: The room the state groups belong to (must all be in the + same room). + state_groups_to_delete (Collection[int]): Set of all state groups + to delete. + """ + + return self.runInteraction( + "purge_unreferenced_state_groups", + self._purge_unreferenced_state_groups, + room_id, + state_groups_to_delete, + ) + + def _purge_unreferenced_state_groups(self, txn, room_id, state_groups_to_delete): + logger.info( + "[purge] found %i state groups to delete", len(state_groups_to_delete) + ) + + rows = self._simple_select_many_txn( + txn, + table="state_group_edges", + column="prev_state_group", + iterable=state_groups_to_delete, + keyvalues={}, + retcols=("state_group",), + ) + + remaining_state_groups = set( + row["state_group"] + for row in rows + if row["state_group"] not in state_groups_to_delete + ) + + logger.info( + "[purge] de-delta-ing %i remaining state groups", + len(remaining_state_groups), + ) + + # Now we turn the state groups that reference to-be-deleted state + # groups to non delta versions. + for sg in remaining_state_groups: + logger.info("[purge] de-delta-ing remaining state group %s", sg) + curr_state = self._get_state_groups_from_groups_txn(txn, [sg]) + curr_state = curr_state[sg] + + self._simple_delete_txn( + txn, table="state_groups_state", keyvalues={"state_group": sg} + ) + + self._simple_delete_txn( + txn, table="state_group_edges", keyvalues={"state_group": sg} + ) + + self._simple_insert_many_txn( + txn, + table="state_groups_state", + values=[ + { + "state_group": sg, + "room_id": room_id, + "type": key[0], + "state_key": key[1], + "event_id": state_id, + } + for key, state_id in iteritems(curr_state) + ], + ) + + logger.info("[purge] removing redundant state groups") + txn.executemany( + "DELETE FROM state_groups_state WHERE state_group = ?", + ((sg,) for sg in state_groups_to_delete), + ) + txn.executemany( + "DELETE FROM state_groups WHERE id = ?", + ((sg,) for sg in state_groups_to_delete), + ) + + @defer.inlineCallbacks + def get_previous_state_groups(self, state_groups): + """Fetch the previous groups of the given state groups. + + Args: + state_groups (Iterable[int]) + + Returns: + Deferred[dict[int, int]]: mapping from state group to previous + state group. + """ + + rows = yield self._simple_select_many_batch( + table="state_group_edges", + column="prev_state_group", + iterable=state_groups, + keyvalues={}, + retcols=("prev_state_group", "state_group"), + desc="get_previous_state_groups", + ) + + return {row["state_group"]: row["prev_state_group"] for row in rows} + + def purge_room_state(self, room_id, state_groups_to_delete): + """Deletes all record of a room from state tables + + Args: + room_id (str): + state_groups_to_delete (list[int]): State groups to delete + """ + + return self.runInteraction( + "purge_room_state", + self._purge_room_state_txn, + room_id, + state_groups_to_delete, + ) + + def _purge_room_state_txn(self, txn, room_id, state_groups_to_delete): + # first we have to delete the state groups states + logger.info("[purge] removing %s from state_groups_state", room_id) + + self._simple_delete_many_txn( + txn, + table="state_groups_state", + column="state_group", + iterable=state_groups_to_delete, + keyvalues={}, + ) + + # ... and the state group edges + logger.info("[purge] removing %s from state_group_edges", room_id) + + self._simple_delete_many_txn( + txn, + table="state_group_edges", + column="state_group", + iterable=state_groups_to_delete, + keyvalues={}, + ) + + # ... and the state groups + logger.info("[purge] removing %s from state_groups", room_id) + + self._simple_delete_many_txn( + txn, + table="state_groups", + column="id", + iterable=state_groups_to_delete, + keyvalues={}, + ) + async def is_event_after(self, event_id1, event_id2): """Returns True if event_id1 is after event_id2 in the stream """ diff --git a/synapse/storage/data_stores/main/events_bg_updates.py b/synapse/storage/data_stores/main/events_bg_updates.py index 51352b9966..0ed59ef48e 100644 --- a/synapse/storage/data_stores/main/events_bg_updates.py +++ b/synapse/storage/data_stores/main/events_bg_updates.py @@ -21,6 +21,7 @@ from canonicaljson import json from twisted.internet import defer +from synapse.api.constants import EventContentFields from synapse.storage._base import make_in_list_sql_clause from synapse.storage.background_updates import BackgroundUpdateStore @@ -85,6 +86,10 @@ class EventsBackgroundUpdatesStore(BackgroundUpdateStore): "event_fix_redactions_bytes", self._event_fix_redactions_bytes ) + self.register_background_update_handler( + "event_store_labels", self._event_store_labels + ) + @defer.inlineCallbacks def _background_reindex_fields_sender(self, progress, batch_size): target_min_stream_id = progress["target_min_stream_id_inclusive"] @@ -503,3 +508,61 @@ class EventsBackgroundUpdatesStore(BackgroundUpdateStore): yield self._end_background_update("event_fix_redactions_bytes") return 1 + + @defer.inlineCallbacks + def _event_store_labels(self, progress, batch_size): + """Background update handler which will store labels for existing events.""" + last_event_id = progress.get("last_event_id", "") + + def _event_store_labels_txn(txn): + txn.execute( + """ + SELECT event_id, json FROM event_json + LEFT JOIN event_labels USING (event_id) + WHERE event_id > ? AND label IS NULL + ORDER BY event_id LIMIT ? + """, + (last_event_id, batch_size), + ) + + results = list(txn) + + nbrows = 0 + last_row_event_id = "" + for (event_id, event_json_raw) in results: + event_json = json.loads(event_json_raw) + + self._simple_insert_many_txn( + txn=txn, + table="event_labels", + values=[ + { + "event_id": event_id, + "label": label, + "room_id": event_json["room_id"], + "topological_ordering": event_json["depth"], + } + for label in event_json["content"].get( + EventContentFields.LABELS, [] + ) + if isinstance(label, str) + ], + ) + + nbrows += 1 + last_row_event_id = event_id + + self._background_update_progress_txn( + txn, "event_store_labels", {"last_event_id": last_row_event_id} + ) + + return nbrows + + num_rows = yield self.runInteraction( + desc="event_store_labels", func=_event_store_labels_txn + ) + + if not num_rows: + yield self._end_background_update("event_store_labels") + + return num_rows diff --git a/synapse/storage/data_stores/main/group_server.py b/synapse/storage/data_stores/main/group_server.py index b3a2771f1b..5ded539af8 100644 --- a/synapse/storage/data_stores/main/group_server.py +++ b/synapse/storage/data_stores/main/group_server.py @@ -553,6 +553,21 @@ class GroupServerStore(SQLBaseStore): desc="remove_user_from_summary", ) + def get_local_groups_for_room(self, room_id): + """Get all of the local group that contain a given room + Args: + room_id (str): The ID of a room + Returns: + Deferred[list[str]]: A twisted.Deferred containing a list of group ids + containing this room + """ + return self._simple_select_onecol( + table="group_rooms", + keyvalues={"room_id": room_id}, + retcol="group_id", + desc="get_local_groups_for_room", + ) + def get_users_for_summary_by_role(self, group_id, include_private=False): """Get the users and roles that should be included in a summary request diff --git a/synapse/storage/data_stores/main/registration.py b/synapse/storage/data_stores/main/registration.py index f70d41ecab..ee1b2b2bbf 100644 --- a/synapse/storage/data_stores/main/registration.py +++ b/synapse/storage/data_stores/main/registration.py @@ -488,14 +488,14 @@ class RegistrationWorkerStore(SQLBaseStore): we can. Unfortunately, it's possible some of them are already taken by existing users, and there may be gaps in the already taken range. This function returns the start of the first allocatable gap. This is to - avoid the case of ID 10000000 being pre-allocated, so us wasting the - first (and shortest) many generated user IDs. + avoid the case of ID 1000 being pre-allocated and starting at 1001 while + 0-999 are available. """ def _find_next_generated_user_id(txn): - # We bound between '@1' and '@a' to avoid pulling the entire table + # We bound between '@0' and '@a' to avoid pulling the entire table # out. - txn.execute("SELECT name FROM users WHERE '@1' <= name AND name < '@a'") + txn.execute("SELECT name FROM users WHERE '@0' <= name AND name < '@a'") regex = re.compile(r"^@(\d+):") diff --git a/synapse/storage/data_stores/main/schema/delta/56/event_labels_background_update.sql b/synapse/storage/data_stores/main/schema/delta/56/event_labels_background_update.sql new file mode 100644 index 0000000000..5f5e0499ae --- /dev/null +++ b/synapse/storage/data_stores/main/schema/delta/56/event_labels_background_update.sql @@ -0,0 +1,17 @@ +/* Copyright 2019 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. + */ + +INSERT INTO background_updates (update_name, progress_json) VALUES + ('event_store_labels', '{}'); diff --git a/synapse/storage/data_stores/main/state.py b/synapse/storage/data_stores/main/state.py index 3132848034..6a90daea31 100644 --- a/synapse/storage/data_stores/main/state.py +++ b/synapse/storage/data_stores/main/state.py @@ -285,7 +285,11 @@ class StateGroupWorkerStore( room_id (str) Returns: - Deferred[unicode|None]: predecessor room id + Deferred[dict|None]: A dictionary containing the structure of the predecessor + field from the room's create event. The structure is subject to other servers, + but it is expected to be: + * room_id (str): The room ID of the predecessor room + * event_id (str): The ID of the tombstone event in the predecessor room Raises: NotFoundError if the room is unknown @@ -991,6 +995,29 @@ class StateGroupWorkerStore( return self.runInteraction("store_state_group", _store_state_group_txn) + @defer.inlineCallbacks + def get_referenced_state_groups(self, state_groups): + """Check if the state groups are referenced by events. + + Args: + state_groups (Iterable[int]) + + Returns: + Deferred[set[int]]: The subset of state groups that are + referenced. + """ + + rows = yield self._simple_select_many_batch( + table="event_to_state_groups", + column="state_group", + iterable=state_groups, + keyvalues={}, + retcols=("DISTINCT state_group",), + desc="get_referenced_state_groups", + ) + + return set(row["state_group"] for row in rows) + class StateBackgroundUpdateStore( StateGroupBackgroundUpdateStore, BackgroundUpdateStore @@ -1231,7 +1258,7 @@ class StateStore(StateGroupWorkerStore, StateBackgroundUpdateStore): # 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 + state_groups[event.event_id] = context.state_group_before_event continue state_groups[event.event_id] = context.state_group diff --git a/synapse/storage/data_stores/main/stream.py b/synapse/storage/data_stores/main/stream.py index 616ef91d4e..8780fdd989 100644 --- a/synapse/storage/data_stores/main/stream.py +++ b/synapse/storage/data_stores/main/stream.py @@ -871,14 +871,38 @@ class StreamWorkerStore(EventsWorkerStore, SQLBaseStore): args.append(int(limit)) - sql = ( - "SELECT DISTINCT event_id, topological_ordering, stream_ordering" - " FROM events" - " LEFT JOIN event_labels USING (event_id, room_id, topological_ordering)" - " WHERE outlier = ? AND room_id = ? AND %(bounds)s" - " ORDER BY topological_ordering %(order)s," - " stream_ordering %(order)s LIMIT ?" - ) % {"bounds": bounds, "order": order} + select_keywords = "SELECT" + join_clause = "" + if event_filter and event_filter.labels: + # If we're not filtering on a label, then joining on event_labels will + # return as many row for a single event as the number of labels it has. To + # avoid this, only join if we're filtering on at least one label. + join_clause = """ + LEFT JOIN event_labels + USING (event_id, room_id, topological_ordering) + """ + if len(event_filter.labels) > 1: + # Using DISTINCT in this SELECT query is quite expensive, because it + # requires the engine to sort on the entire (not limited) result set, + # i.e. the entire events table. We only need to use it when we're + # filtering on more than two labels, because that's the only scenario + # in which we can possibly to get multiple times the same event ID in + # the results. + select_keywords += "DISTINCT" + + sql = """ + %(select_keywords)s event_id, topological_ordering, stream_ordering + FROM events + %(join_clause)s + WHERE outlier = ? AND room_id = ? AND %(bounds)s + ORDER BY topological_ordering %(order)s, + stream_ordering %(order)s LIMIT ? + """ % { + "select_keywords": select_keywords, + "join_clause": join_clause, + "bounds": bounds, + "order": order, + } txn.execute(sql, args) diff --git a/synapse/storage/purge_events.py b/synapse/storage/purge_events.py new file mode 100644 index 0000000000..a368182034 --- /dev/null +++ b/synapse/storage/purge_events.py @@ -0,0 +1,117 @@ +# -*- coding: utf-8 -*- +# Copyright 2019 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 itertools +import logging + +from twisted.internet import defer + +logger = logging.getLogger(__name__) + + +class PurgeEventsStorage(object): + """High level interface for purging rooms and event history. + """ + + def __init__(self, hs, stores): + self.stores = stores + + @defer.inlineCallbacks + def purge_room(self, room_id: str): + """Deletes all record of a room + """ + + state_groups_to_delete = yield self.stores.main.purge_room(room_id) + yield self.stores.main.purge_room_state(room_id, state_groups_to_delete) + + @defer.inlineCallbacks + def purge_history(self, room_id, token, delete_local_events): + """Deletes room history before a certain point + + Args: + room_id (str): + + token (str): A topological token to delete events before + + delete_local_events (bool): + if True, we will delete local events as well as remote ones + (instead of just marking them as outliers and deleting their + state groups). + """ + state_groups = yield self.stores.main.purge_history( + room_id, token, delete_local_events + ) + + logger.info("[purge] finding state groups that can be deleted") + + sg_to_delete = yield self._find_unreferenced_groups(state_groups) + + yield self.stores.main.purge_unreferenced_state_groups(room_id, sg_to_delete) + + @defer.inlineCallbacks + def _find_unreferenced_groups(self, state_groups): + """Used when purging history to figure out which state groups can be + deleted. + + Args: + state_groups (set[int]): Set of state groups referenced by events + that are going to be deleted. + + Returns: + Deferred[set[int]] The set of state groups that can be deleted. + """ + # Graph of state group -> previous group + graph = {} + + # Set of events that we have found to be referenced by events + referenced_groups = set() + + # Set of state groups we've already seen + state_groups_seen = set(state_groups) + + # Set of state groups to handle next. + next_to_search = set(state_groups) + while next_to_search: + # We bound size of groups we're looking up at once, to stop the + # SQL query getting too big + if len(next_to_search) < 100: + current_search = next_to_search + next_to_search = set() + else: + current_search = set(itertools.islice(next_to_search, 100)) + next_to_search -= current_search + + referenced = yield self.stores.main.get_referenced_state_groups( + current_search + ) + referenced_groups |= referenced + + # We don't continue iterating up the state group graphs for state + # groups that are referenced. + current_search -= referenced + + edges = yield self.stores.main.get_previous_state_groups(current_search) + + prevs = set(edges.values()) + # We don't bother re-handling groups we've already seen + prevs -= state_groups_seen + next_to_search |= prevs + state_groups_seen |= prevs + + graph.update(edges) + + to_delete = state_groups_seen - referenced_groups + + return to_delete diff --git a/synapse/util/caches/descriptors.py b/synapse/util/caches/descriptors.py index 0e8da27f53..84f5ae22c3 100644 --- a/synapse/util/caches/descriptors.py +++ b/synapse/util/caches/descriptors.py @@ -17,8 +17,8 @@ import functools import inspect import logging import threading -from collections import namedtuple -from typing import Any, cast +from typing import Any, Tuple, Union, cast +from weakref import WeakValueDictionary from six import itervalues @@ -38,6 +38,8 @@ from . import register_cache logger = logging.getLogger(__name__) +CacheKey = Union[Tuple, Any] + class _CachedFunction(Protocol): invalidate = None # type: Any @@ -430,7 +432,7 @@ class CacheDescriptor(_CacheDescriptorBase): # 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(cache, cache_key) + kwargs["cache_context"] = _CacheContext.get_instance(cache, cache_key) try: cached_result_d = cache.get(cache_key, callback=invalidate_callback) @@ -624,14 +626,38 @@ class CacheListDescriptor(_CacheDescriptorBase): return wrapped -class _CacheContext(namedtuple("_CacheContext", ("cache", "key"))): - # We rely on _CacheContext implementing __eq__ and __hash__ sensibly, - # which namedtuple does for us (i.e. two _CacheContext are the same if - # their caches and keys match). This is important in particular to - # dedupe when we add callbacks to lru cache nodes, otherwise the number - # of callbacks would grow. - def invalidate(self): - self.cache.invalidate(self.key) +class _CacheContext: + """Holds cache information from the cached function higher in the calling order. + + Can be used to invalidate the higher level cache entry if something changes + on a lower level. + """ + + _cache_context_objects = ( + WeakValueDictionary() + ) # type: WeakValueDictionary[Tuple[Cache, CacheKey], _CacheContext] + + def __init__(self, cache, cache_key): # type: (Cache, CacheKey) -> None + self._cache = cache + self._cache_key = cache_key + + def invalidate(self): # type: () -> None + """Invalidates the cache entry referred to by the context.""" + self._cache.invalidate(self._cache_key) + + @classmethod + def get_instance(cls, cache, cache_key): # type: (Cache, CacheKey) -> _CacheContext + """Returns an instance constructed with the given arguments. + + A new instance is only created if none already exists. + """ + + # We make sure there are no identical _CacheContext instances. This is + # important in particular to dedupe when we add callbacks to lru cache + # nodes, otherwise the number of callbacks would grow. + return cls._cache_context_objects.setdefault( + (cache, cache_key), cls(cache, cache_key) + ) def cached( diff --git a/synapse/util/httpresourcetree.py b/synapse/util/httpresourcetree.py index 1a20c596bf..3c0e8469f3 100644 --- a/synapse/util/httpresourcetree.py +++ b/synapse/util/httpresourcetree.py @@ -20,7 +20,7 @@ logger = logging.getLogger(__name__) def create_resource_tree(desired_tree, root_resource): - """Create the resource tree for this Home Server. + """Create the resource tree for this homeserver. This in unduly complicated because Twisted does not support putting child resources more than 1 level deep at a time. |