diff options
Diffstat (limited to 'synapse')
142 files changed, 7601 insertions, 4004 deletions
diff --git a/synapse/__init__.py b/synapse/__init__.py index 80efad3a65..a9d5198aba 100644 --- a/synapse/__init__.py +++ b/synapse/__init__.py @@ -16,4 +16,4 @@ """ This is a reference implementation of a Matrix home server. """ -__version__ = "0.26.1" +__version__ = "0.27.2" diff --git a/synapse/api/errors.py b/synapse/api/errors.py index 79b35b3e7c..aa15f73f36 100644 --- a/synapse/api/errors.py +++ b/synapse/api/errors.py @@ -46,6 +46,7 @@ class Codes(object): THREEPID_AUTH_FAILED = "M_THREEPID_AUTH_FAILED" THREEPID_IN_USE = "M_THREEPID_IN_USE" THREEPID_NOT_FOUND = "M_THREEPID_NOT_FOUND" + THREEPID_DENIED = "M_THREEPID_DENIED" INVALID_USERNAME = "M_INVALID_USERNAME" SERVER_NOT_TRUSTED = "M_SERVER_NOT_TRUSTED" @@ -140,6 +141,32 @@ class RegistrationError(SynapseError): pass +class FederationDeniedError(SynapseError): + """An error raised when the server tries to federate with a server which + is not on its federation whitelist. + + Attributes: + destination (str): The destination which has been denied + """ + + def __init__(self, destination): + """Raised by federation client or server to indicate that we are + are deliberately not attempting to contact a given server because it is + not on our federation whitelist. + + Args: + destination (str): the domain in question + """ + + self.destination = destination + + super(FederationDeniedError, self).__init__( + code=403, + msg="Federation denied with %s." % (self.destination,), + errcode=Codes.FORBIDDEN, + ) + + class InteractiveAuthIncompleteError(Exception): """An error raised when UI auth is not yet complete diff --git a/synapse/app/_base.py b/synapse/app/_base.py index 9477737759..e4318cdfc3 100644 --- a/synapse/app/_base.py +++ b/synapse/app/_base.py @@ -25,7 +25,9 @@ except Exception: from daemonize import Daemonize from synapse.util import PreserveLoggingContext from synapse.util.rlimit import change_resource_limit -from twisted.internet import reactor +from twisted.internet import error, reactor + +logger = logging.getLogger(__name__) def start_worker_reactor(appname, config): @@ -120,3 +122,57 @@ def quit_with_error(error_string): sys.stderr.write(" %s\n" % (line.rstrip(),)) sys.stderr.write("*" * line_length + '\n') sys.exit(1) + + +def listen_tcp(bind_addresses, port, factory, backlog=50): + """ + Create a TCP socket for a port and several addresses + """ + for address in bind_addresses: + try: + reactor.listenTCP( + port, + factory, + backlog, + address + ) + except error.CannotListenError as e: + check_bind_error(e, address, bind_addresses) + + +def listen_ssl(bind_addresses, port, factory, context_factory, backlog=50): + """ + Create an SSL socket for a port and several addresses + """ + for address in bind_addresses: + try: + reactor.listenSSL( + port, + factory, + context_factory, + backlog, + address + ) + except error.CannotListenError as e: + check_bind_error(e, address, bind_addresses) + + +def check_bind_error(e, address, bind_addresses): + """ + This method checks an exception occurred while binding on 0.0.0.0. + If :: is specified in the bind addresses a warning is shown. + The exception is still raised otherwise. + + Binding on both 0.0.0.0 and :: causes an exception on Linux and macOS + because :: binds on both IPv4 and IPv6 (as per RFC 3493). + When binding on 0.0.0.0 after :: this can safely be ignored. + + Args: + e (Exception): Exception that was caught. + address (str): Address on which binding was attempted. + bind_addresses (list): Addresses on which the service listens. + """ + if address == '0.0.0.0' and '::' in bind_addresses: + logger.warn('Failed to listen on 0.0.0.0, continuing because listening on [::]') + else: + raise e diff --git a/synapse/app/appservice.py b/synapse/app/appservice.py index ba2657bbad..c6fe4516d1 100644 --- a/synapse/app/appservice.py +++ b/synapse/app/appservice.py @@ -49,19 +49,6 @@ class AppserviceSlaveStore( class AppserviceServer(HomeServer): - def get_db_conn(self, run_new_connection=True): - # Any param beginning with cp_ is a parameter for adbapi, and should - # not be passed to the database engine. - db_params = { - k: v for k, v in self.db_config.get("args", {}).items() - if not k.startswith("cp_") - } - db_conn = self.database_engine.module.connect(**db_params) - - if run_new_connection: - self.database_engine.on_new_connection(db_conn) - return db_conn - def setup(self): logger.info("Setting up.") self.datastore = AppserviceSlaveStore(self.get_db_conn(), self) @@ -79,17 +66,16 @@ class AppserviceServer(HomeServer): root_resource = create_resource_tree(resources, Resource()) - for address in bind_addresses: - reactor.listenTCP( - port, - SynapseSite( - "synapse.access.http.%s" % (site_tag,), - site_tag, - listener_config, - root_resource, - ), - interface=address + _base.listen_tcp( + bind_addresses, + port, + SynapseSite( + "synapse.access.http.%s" % (site_tag,), + site_tag, + listener_config, + root_resource, ) + ) logger.info("Synapse appservice now listening on port %d", port) @@ -98,18 +84,15 @@ class AppserviceServer(HomeServer): if listener["type"] == "http": self._listen_http(listener) elif listener["type"] == "manhole": - bind_addresses = listener["bind_addresses"] - - for address in bind_addresses: - reactor.listenTCP( - listener["port"], - manhole( - username="matrix", - password="rabbithole", - globals={"hs": self}, - ), - interface=address + _base.listen_tcp( + listener["bind_addresses"], + listener["port"], + manhole( + username="matrix", + password="rabbithole", + globals={"hs": self}, ) + ) else: logger.warn("Unrecognized listener type: %s", listener["type"]) diff --git a/synapse/app/client_reader.py b/synapse/app/client_reader.py index 129cfa901f..0a8ce9bc66 100644 --- a/synapse/app/client_reader.py +++ b/synapse/app/client_reader.py @@ -64,19 +64,6 @@ class ClientReaderSlavedStore( class ClientReaderServer(HomeServer): - def get_db_conn(self, run_new_connection=True): - # Any param beginning with cp_ is a parameter for adbapi, and should - # not be passed to the database engine. - db_params = { - k: v for k, v in self.db_config.get("args", {}).items() - if not k.startswith("cp_") - } - db_conn = self.database_engine.module.connect(**db_params) - - if run_new_connection: - self.database_engine.on_new_connection(db_conn) - return db_conn - def setup(self): logger.info("Setting up.") self.datastore = ClientReaderSlavedStore(self.get_db_conn(), self) @@ -103,17 +90,16 @@ class ClientReaderServer(HomeServer): root_resource = create_resource_tree(resources, Resource()) - for address in bind_addresses: - reactor.listenTCP( - port, - SynapseSite( - "synapse.access.http.%s" % (site_tag,), - site_tag, - listener_config, - root_resource, - ), - interface=address + _base.listen_tcp( + bind_addresses, + port, + SynapseSite( + "synapse.access.http.%s" % (site_tag,), + site_tag, + listener_config, + root_resource, ) + ) logger.info("Synapse client reader now listening on port %d", port) @@ -122,18 +108,16 @@ class ClientReaderServer(HomeServer): if listener["type"] == "http": self._listen_http(listener) elif listener["type"] == "manhole": - bind_addresses = listener["bind_addresses"] - - for address in bind_addresses: - reactor.listenTCP( - listener["port"], - manhole( - username="matrix", - password="rabbithole", - globals={"hs": self}, - ), - interface=address + _base.listen_tcp( + listener["bind_addresses"], + listener["port"], + manhole( + username="matrix", + password="rabbithole", + globals={"hs": self}, ) + ) + else: logger.warn("Unrecognized listener type: %s", listener["type"]) @@ -172,7 +156,6 @@ def start(config_options): ) ss.setup() - ss.get_handlers() ss.start_listening(config.worker_listeners) def start(): diff --git a/synapse/app/event_creator.py b/synapse/app/event_creator.py new file mode 100644 index 0000000000..172e989b54 --- /dev/null +++ b/synapse/app/event_creator.py @@ -0,0 +1,189 @@ +#!/usr/bin/env python +# -*- coding: utf-8 -*- +# Copyright 2018 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. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. +import logging +import sys + +import synapse +from synapse import events +from synapse.app import _base +from synapse.config._base import ConfigError +from synapse.config.homeserver import HomeServerConfig +from synapse.config.logger import setup_logging +from synapse.crypto import context_factory +from synapse.http.server import JsonResource +from synapse.http.site import SynapseSite +from synapse.metrics.resource import METRICS_PREFIX, MetricsResource +from synapse.replication.slave.storage._base import BaseSlavedStore +from synapse.replication.slave.storage.account_data import SlavedAccountDataStore +from synapse.replication.slave.storage.appservice import SlavedApplicationServiceStore +from synapse.replication.slave.storage.client_ips import SlavedClientIpStore +from synapse.replication.slave.storage.devices import SlavedDeviceStore +from synapse.replication.slave.storage.directory import DirectoryStore +from synapse.replication.slave.storage.events import SlavedEventStore +from synapse.replication.slave.storage.profile import SlavedProfileStore +from synapse.replication.slave.storage.push_rule import SlavedPushRuleStore +from synapse.replication.slave.storage.pushers import SlavedPusherStore +from synapse.replication.slave.storage.receipts import SlavedReceiptsStore +from synapse.replication.slave.storage.registration import SlavedRegistrationStore +from synapse.replication.slave.storage.room import RoomStore +from synapse.replication.slave.storage.transactions import TransactionStore +from synapse.replication.tcp.client import ReplicationClientHandler +from synapse.rest.client.v1.room import ( + RoomSendEventRestServlet, RoomMembershipRestServlet, RoomStateEventRestServlet, + JoinRoomAliasServlet, +) +from synapse.server import HomeServer +from synapse.storage.engines import create_engine +from synapse.util.httpresourcetree import create_resource_tree +from synapse.util.logcontext import LoggingContext +from synapse.util.manhole import manhole +from synapse.util.versionstring import get_version_string +from twisted.internet import reactor +from twisted.web.resource import Resource + +logger = logging.getLogger("synapse.app.event_creator") + + +class EventCreatorSlavedStore( + DirectoryStore, + TransactionStore, + SlavedProfileStore, + SlavedAccountDataStore, + SlavedPusherStore, + SlavedReceiptsStore, + SlavedPushRuleStore, + SlavedDeviceStore, + SlavedClientIpStore, + SlavedApplicationServiceStore, + SlavedEventStore, + SlavedRegistrationStore, + RoomStore, + BaseSlavedStore, +): + pass + + +class EventCreatorServer(HomeServer): + def setup(self): + logger.info("Setting up.") + self.datastore = EventCreatorSlavedStore(self.get_db_conn(), self) + logger.info("Finished setting up.") + + def _listen_http(self, listener_config): + port = listener_config["port"] + bind_addresses = listener_config["bind_addresses"] + site_tag = listener_config.get("tag", port) + resources = {} + for res in listener_config["resources"]: + for name in res["names"]: + if name == "metrics": + resources[METRICS_PREFIX] = MetricsResource(self) + elif name == "client": + resource = JsonResource(self, canonical_json=False) + RoomSendEventRestServlet(self).register(resource) + RoomMembershipRestServlet(self).register(resource) + RoomStateEventRestServlet(self).register(resource) + JoinRoomAliasServlet(self).register(resource) + resources.update({ + "/_matrix/client/r0": resource, + "/_matrix/client/unstable": resource, + "/_matrix/client/v2_alpha": resource, + "/_matrix/client/api/v1": resource, + }) + + root_resource = create_resource_tree(resources, Resource()) + + _base.listen_tcp( + bind_addresses, + port, + SynapseSite( + "synapse.access.http.%s" % (site_tag,), + site_tag, + listener_config, + root_resource, + ) + ) + + logger.info("Synapse event creator now listening on port %d", port) + + def start_listening(self, listeners): + for listener in listeners: + if listener["type"] == "http": + self._listen_http(listener) + elif listener["type"] == "manhole": + _base.listen_tcp( + listener["bind_addresses"], + listener["port"], + manhole( + username="matrix", + password="rabbithole", + globals={"hs": self}, + ) + ) + else: + logger.warn("Unrecognized listener type: %s", listener["type"]) + + self.get_tcp_replication().start_replication(self) + + def build_tcp_replication(self): + return ReplicationClientHandler(self.get_datastore()) + + +def start(config_options): + try: + config = HomeServerConfig.load_config( + "Synapse event creator", config_options + ) + except ConfigError as e: + sys.stderr.write("\n" + e.message + "\n") + sys.exit(1) + + assert config.worker_app == "synapse.app.event_creator" + + assert config.worker_replication_http_port is not None + + setup_logging(config, use_worker_options=True) + + events.USE_FROZEN_DICTS = config.use_frozen_dicts + + database_engine = create_engine(config.database_config) + + tls_server_context_factory = context_factory.ServerContextFactory(config) + + ss = EventCreatorServer( + config.server_name, + db_config=config.database_config, + tls_server_context_factory=tls_server_context_factory, + config=config, + version_string="Synapse/" + get_version_string(synapse), + database_engine=database_engine, + ) + + ss.setup() + ss.start_listening(config.worker_listeners) + + def start(): + ss.get_state_handler().start_caching() + ss.get_datastore().start_profiling() + + reactor.callWhenRunning(start) + + _base.start_worker_reactor("synapse-event-creator", config) + + +if __name__ == '__main__': + with LoggingContext("main"): + start(sys.argv[1:]) diff --git a/synapse/app/federation_reader.py b/synapse/app/federation_reader.py index 40cebe6f4a..20d157911b 100644 --- a/synapse/app/federation_reader.py +++ b/synapse/app/federation_reader.py @@ -58,19 +58,6 @@ class FederationReaderSlavedStore( class FederationReaderServer(HomeServer): - def get_db_conn(self, run_new_connection=True): - # Any param beginning with cp_ is a parameter for adbapi, and should - # not be passed to the database engine. - db_params = { - k: v for k, v in self.db_config.get("args", {}).items() - if not k.startswith("cp_") - } - db_conn = self.database_engine.module.connect(**db_params) - - if run_new_connection: - self.database_engine.on_new_connection(db_conn) - return db_conn - def setup(self): logger.info("Setting up.") self.datastore = FederationReaderSlavedStore(self.get_db_conn(), self) @@ -92,17 +79,16 @@ class FederationReaderServer(HomeServer): root_resource = create_resource_tree(resources, Resource()) - for address in bind_addresses: - reactor.listenTCP( - port, - SynapseSite( - "synapse.access.http.%s" % (site_tag,), - site_tag, - listener_config, - root_resource, - ), - interface=address + _base.listen_tcp( + bind_addresses, + port, + SynapseSite( + "synapse.access.http.%s" % (site_tag,), + site_tag, + listener_config, + root_resource, ) + ) logger.info("Synapse federation reader now listening on port %d", port) @@ -111,18 +97,15 @@ class FederationReaderServer(HomeServer): if listener["type"] == "http": self._listen_http(listener) elif listener["type"] == "manhole": - bind_addresses = listener["bind_addresses"] - - for address in bind_addresses: - reactor.listenTCP( - listener["port"], - manhole( - username="matrix", - password="rabbithole", - globals={"hs": self}, - ), - interface=address + _base.listen_tcp( + listener["bind_addresses"], + listener["port"], + manhole( + username="matrix", + password="rabbithole", + globals={"hs": self}, ) + ) else: logger.warn("Unrecognized listener type: %s", listener["type"]) @@ -161,7 +144,6 @@ def start(config_options): ) ss.setup() - ss.get_handlers() ss.start_listening(config.worker_listeners) def start(): diff --git a/synapse/app/federation_sender.py b/synapse/app/federation_sender.py index 389e3909d1..f760826d27 100644 --- a/synapse/app/federation_sender.py +++ b/synapse/app/federation_sender.py @@ -76,19 +76,6 @@ class FederationSenderSlaveStore( class FederationSenderServer(HomeServer): - def get_db_conn(self, run_new_connection=True): - # Any param beginning with cp_ is a parameter for adbapi, and should - # not be passed to the database engine. - db_params = { - k: v for k, v in self.db_config.get("args", {}).items() - if not k.startswith("cp_") - } - db_conn = self.database_engine.module.connect(**db_params) - - if run_new_connection: - self.database_engine.on_new_connection(db_conn) - return db_conn - def setup(self): logger.info("Setting up.") self.datastore = FederationSenderSlaveStore(self.get_db_conn(), self) @@ -106,17 +93,16 @@ class FederationSenderServer(HomeServer): root_resource = create_resource_tree(resources, Resource()) - for address in bind_addresses: - reactor.listenTCP( - port, - SynapseSite( - "synapse.access.http.%s" % (site_tag,), - site_tag, - listener_config, - root_resource, - ), - interface=address + _base.listen_tcp( + bind_addresses, + port, + SynapseSite( + "synapse.access.http.%s" % (site_tag,), + site_tag, + listener_config, + root_resource, ) + ) logger.info("Synapse federation_sender now listening on port %d", port) @@ -125,18 +111,15 @@ class FederationSenderServer(HomeServer): if listener["type"] == "http": self._listen_http(listener) elif listener["type"] == "manhole": - bind_addresses = listener["bind_addresses"] - - for address in bind_addresses: - reactor.listenTCP( - listener["port"], - manhole( - username="matrix", - password="rabbithole", - globals={"hs": self}, - ), - interface=address + _base.listen_tcp( + listener["bind_addresses"], + listener["port"], + manhole( + username="matrix", + password="rabbithole", + globals={"hs": self}, ) + ) else: logger.warn("Unrecognized listener type: %s", listener["type"]) diff --git a/synapse/app/frontend_proxy.py b/synapse/app/frontend_proxy.py index abc7ef5725..816c080d18 100644 --- a/synapse/app/frontend_proxy.py +++ b/synapse/app/frontend_proxy.py @@ -118,19 +118,6 @@ class FrontendProxySlavedStore( class FrontendProxyServer(HomeServer): - def get_db_conn(self, run_new_connection=True): - # Any param beginning with cp_ is a parameter for adbapi, and should - # not be passed to the database engine. - db_params = { - k: v for k, v in self.db_config.get("args", {}).items() - if not k.startswith("cp_") - } - db_conn = self.database_engine.module.connect(**db_params) - - if run_new_connection: - self.database_engine.on_new_connection(db_conn) - return db_conn - def setup(self): logger.info("Setting up.") self.datastore = FrontendProxySlavedStore(self.get_db_conn(), self) @@ -157,17 +144,16 @@ class FrontendProxyServer(HomeServer): root_resource = create_resource_tree(resources, Resource()) - for address in bind_addresses: - reactor.listenTCP( - port, - SynapseSite( - "synapse.access.http.%s" % (site_tag,), - site_tag, - listener_config, - root_resource, - ), - interface=address + _base.listen_tcp( + bind_addresses, + port, + SynapseSite( + "synapse.access.http.%s" % (site_tag,), + site_tag, + listener_config, + root_resource, ) + ) logger.info("Synapse client reader now listening on port %d", port) @@ -176,18 +162,15 @@ class FrontendProxyServer(HomeServer): if listener["type"] == "http": self._listen_http(listener) elif listener["type"] == "manhole": - bind_addresses = listener["bind_addresses"] - - for address in bind_addresses: - reactor.listenTCP( - listener["port"], - manhole( - username="matrix", - password="rabbithole", - globals={"hs": self}, - ), - interface=address + _base.listen_tcp( + listener["bind_addresses"], + listener["port"], + manhole( + username="matrix", + password="rabbithole", + globals={"hs": self}, ) + ) else: logger.warn("Unrecognized listener type: %s", listener["type"]) @@ -228,7 +211,6 @@ def start(config_options): ) ss.setup() - ss.get_handlers() ss.start_listening(config.worker_listeners) def start(): diff --git a/synapse/app/homeserver.py b/synapse/app/homeserver.py index 6b8875afb4..e477c7ced6 100755 --- a/synapse/app/homeserver.py +++ b/synapse/app/homeserver.py @@ -25,7 +25,7 @@ from synapse.api.urls import CONTENT_REPO_PREFIX, FEDERATION_PREFIX, \ LEGACY_MEDIA_PREFIX, MEDIA_PREFIX, SERVER_KEY_PREFIX, SERVER_KEY_V2_PREFIX, \ STATIC_PREFIX, WEB_CLIENT_PREFIX from synapse.app import _base -from synapse.app._base import quit_with_error +from synapse.app._base import quit_with_error, listen_ssl, listen_tcp from synapse.config._base import ConfigError from synapse.config.homeserver import HomeServerConfig from synapse.crypto import context_factory @@ -38,6 +38,7 @@ from synapse.metrics import register_memory_metrics from synapse.metrics.resource import METRICS_PREFIX, MetricsResource from synapse.python_dependencies import CONDITIONAL_REQUIREMENTS, \ check_requirements +from synapse.replication.http import ReplicationRestResource, REPLICATION_PREFIX from synapse.replication.tcp.resource import ReplicationStreamProtocolFactory from synapse.rest import ClientRestResource from synapse.rest.key.v1.server_key_resource import LocalKey @@ -130,30 +131,29 @@ class SynapseHomeServer(HomeServer): root_resource = create_resource_tree(resources, root_resource) if tls: - for address in bind_addresses: - reactor.listenSSL( - port, - SynapseSite( - "synapse.access.https.%s" % (site_tag,), - site_tag, - listener_config, - root_resource, - ), - self.tls_server_context_factory, - interface=address - ) + listen_ssl( + bind_addresses, + port, + SynapseSite( + "synapse.access.https.%s" % (site_tag,), + site_tag, + listener_config, + root_resource, + ), + self.tls_server_context_factory, + ) + else: - for address in bind_addresses: - reactor.listenTCP( - port, - SynapseSite( - "synapse.access.http.%s" % (site_tag,), - site_tag, - listener_config, - root_resource, - ), - interface=address + listen_tcp( + bind_addresses, + port, + SynapseSite( + "synapse.access.http.%s" % (site_tag,), + site_tag, + listener_config, + root_resource, ) + ) logger.info("Synapse now listening on port %d", port) def _configure_named_resource(self, name, compress=False): @@ -220,6 +220,9 @@ class SynapseHomeServer(HomeServer): if name == "metrics" and self.get_config().enable_metrics: resources[METRICS_PREFIX] = MetricsResource(self) + if name == "replication": + resources[REPLICATION_PREFIX] = ReplicationRestResource(self) + return resources def start_listening(self): @@ -229,18 +232,15 @@ class SynapseHomeServer(HomeServer): if listener["type"] == "http": self._listener_http(config, listener) elif listener["type"] == "manhole": - bind_addresses = listener["bind_addresses"] - - for address in bind_addresses: - reactor.listenTCP( - listener["port"], - manhole( - username="matrix", - password="rabbithole", - globals={"hs": self}, - ), - interface=address + listen_tcp( + listener["bind_addresses"], + listener["port"], + manhole( + username="matrix", + password="rabbithole", + globals={"hs": self}, ) + ) elif listener["type"] == "replication": bind_addresses = listener["bind_addresses"] for address in bind_addresses: @@ -270,19 +270,6 @@ class SynapseHomeServer(HomeServer): except IncorrectDatabaseSetup as e: quit_with_error(e.message) - def get_db_conn(self, run_new_connection=True): - # Any param beginning with cp_ is a parameter for adbapi, and should - # not be passed to the database engine. - db_params = { - k: v for k, v in self.db_config.get("args", {}).items() - if not k.startswith("cp_") - } - db_conn = self.database_engine.module.connect(**db_params) - - if run_new_connection: - self.database_engine.on_new_connection(db_conn) - return db_conn - def setup(config_options): """ @@ -361,7 +348,7 @@ def setup(config_options): hs.get_state_handler().start_caching() hs.get_datastore().start_profiling() hs.get_datastore().start_doing_background_updates() - hs.get_replication_layer().start_get_pdu_cache() + hs.get_federation_client().start_get_pdu_cache() register_memory_metrics(hs) diff --git a/synapse/app/media_repository.py b/synapse/app/media_repository.py index c4e5f0965d..84c5791b3b 100644 --- a/synapse/app/media_repository.py +++ b/synapse/app/media_repository.py @@ -60,19 +60,6 @@ class MediaRepositorySlavedStore( class MediaRepositoryServer(HomeServer): - def get_db_conn(self, run_new_connection=True): - # Any param beginning with cp_ is a parameter for adbapi, and should - # not be passed to the database engine. - db_params = { - k: v for k, v in self.db_config.get("args", {}).items() - if not k.startswith("cp_") - } - db_conn = self.database_engine.module.connect(**db_params) - - if run_new_connection: - self.database_engine.on_new_connection(db_conn) - return db_conn - def setup(self): logger.info("Setting up.") self.datastore = MediaRepositorySlavedStore(self.get_db_conn(), self) @@ -99,17 +86,16 @@ class MediaRepositoryServer(HomeServer): root_resource = create_resource_tree(resources, Resource()) - for address in bind_addresses: - reactor.listenTCP( - port, - SynapseSite( - "synapse.access.http.%s" % (site_tag,), - site_tag, - listener_config, - root_resource, - ), - interface=address + _base.listen_tcp( + bind_addresses, + port, + SynapseSite( + "synapse.access.http.%s" % (site_tag,), + site_tag, + listener_config, + root_resource, ) + ) logger.info("Synapse media repository now listening on port %d", port) @@ -118,18 +104,15 @@ class MediaRepositoryServer(HomeServer): if listener["type"] == "http": self._listen_http(listener) elif listener["type"] == "manhole": - bind_addresses = listener["bind_addresses"] - - for address in bind_addresses: - reactor.listenTCP( - listener["port"], - manhole( - username="matrix", - password="rabbithole", - globals={"hs": self}, - ), - interface=address + _base.listen_tcp( + listener["bind_addresses"], + listener["port"], + manhole( + username="matrix", + password="rabbithole", + globals={"hs": self}, ) + ) else: logger.warn("Unrecognized listener type: %s", listener["type"]) @@ -175,7 +158,6 @@ def start(config_options): ) ss.setup() - ss.get_handlers() ss.start_listening(config.worker_listeners) def start(): diff --git a/synapse/app/pusher.py b/synapse/app/pusher.py index db9a4d16f4..98a4a7c62c 100644 --- a/synapse/app/pusher.py +++ b/synapse/app/pusher.py @@ -32,7 +32,6 @@ from synapse.replication.tcp.client import ReplicationClientHandler from synapse.server import HomeServer from synapse.storage import DataStore from synapse.storage.engines import create_engine -from synapse.storage.roommember import RoomMemberStore from synapse.util.httpresourcetree import create_resource_tree from synapse.util.logcontext import LoggingContext, preserve_fn from synapse.util.manhole import manhole @@ -75,25 +74,8 @@ class PusherSlaveStore( DataStore.get_profile_displayname.__func__ ) - who_forgot_in_room = ( - RoomMemberStore.__dict__["who_forgot_in_room"] - ) - class PusherServer(HomeServer): - def get_db_conn(self, run_new_connection=True): - # Any param beginning with cp_ is a parameter for adbapi, and should - # not be passed to the database engine. - db_params = { - k: v for k, v in self.db_config.get("args", {}).items() - if not k.startswith("cp_") - } - db_conn = self.database_engine.module.connect(**db_params) - - if run_new_connection: - self.database_engine.on_new_connection(db_conn) - return db_conn - def setup(self): logger.info("Setting up.") self.datastore = PusherSlaveStore(self.get_db_conn(), self) @@ -114,17 +96,16 @@ class PusherServer(HomeServer): root_resource = create_resource_tree(resources, Resource()) - for address in bind_addresses: - reactor.listenTCP( - port, - SynapseSite( - "synapse.access.http.%s" % (site_tag,), - site_tag, - listener_config, - root_resource, - ), - interface=address + _base.listen_tcp( + bind_addresses, + port, + SynapseSite( + "synapse.access.http.%s" % (site_tag,), + site_tag, + listener_config, + root_resource, ) + ) logger.info("Synapse pusher now listening on port %d", port) @@ -133,18 +114,15 @@ class PusherServer(HomeServer): if listener["type"] == "http": self._listen_http(listener) elif listener["type"] == "manhole": - bind_addresses = listener["bind_addresses"] - - for address in bind_addresses: - reactor.listenTCP( - listener["port"], - manhole( - username="matrix", - password="rabbithole", - globals={"hs": self}, - ), - interface=address + _base.listen_tcp( + listener["bind_addresses"], + listener["port"], + manhole( + username="matrix", + password="rabbithole", + globals={"hs": self}, ) + ) else: logger.warn("Unrecognized listener type: %s", listener["type"]) diff --git a/synapse/app/synchrotron.py b/synapse/app/synchrotron.py index 323fddee21..abe91dcfbd 100644 --- a/synapse/app/synchrotron.py +++ b/synapse/app/synchrotron.py @@ -62,8 +62,6 @@ logger = logging.getLogger("synapse.app.synchrotron") class SynchrotronSlavedStore( - SlavedPushRuleStore, - SlavedEventStore, SlavedReceiptsStore, SlavedAccountDataStore, SlavedApplicationServiceStore, @@ -73,14 +71,12 @@ class SynchrotronSlavedStore( SlavedGroupServerStore, SlavedDeviceInboxStore, SlavedDeviceStore, + SlavedPushRuleStore, + SlavedEventStore, SlavedClientIpStore, RoomStore, BaseSlavedStore, ): - who_forgot_in_room = ( - RoomMemberStore.__dict__["who_forgot_in_room"] - ) - did_forget = ( RoomMemberStore.__dict__["did_forget"] ) @@ -246,19 +242,6 @@ class SynchrotronApplicationService(object): class SynchrotronServer(HomeServer): - def get_db_conn(self, run_new_connection=True): - # Any param beginning with cp_ is a parameter for adbapi, and should - # not be passed to the database engine. - db_params = { - k: v for k, v in self.db_config.get("args", {}).items() - if not k.startswith("cp_") - } - db_conn = self.database_engine.module.connect(**db_params) - - if run_new_connection: - self.database_engine.on_new_connection(db_conn) - return db_conn - def setup(self): logger.info("Setting up.") self.datastore = SynchrotronSlavedStore(self.get_db_conn(), self) @@ -288,17 +271,16 @@ class SynchrotronServer(HomeServer): root_resource = create_resource_tree(resources, Resource()) - for address in bind_addresses: - reactor.listenTCP( - port, - SynapseSite( - "synapse.access.http.%s" % (site_tag,), - site_tag, - listener_config, - root_resource, - ), - interface=address + _base.listen_tcp( + bind_addresses, + port, + SynapseSite( + "synapse.access.http.%s" % (site_tag,), + site_tag, + listener_config, + root_resource, ) + ) logger.info("Synapse synchrotron now listening on port %d", port) @@ -307,18 +289,15 @@ class SynchrotronServer(HomeServer): if listener["type"] == "http": self._listen_http(listener) elif listener["type"] == "manhole": - bind_addresses = listener["bind_addresses"] - - for address in bind_addresses: - reactor.listenTCP( - listener["port"], - manhole( - username="matrix", - password="rabbithole", - globals={"hs": self}, - ), - interface=address + _base.listen_tcp( + listener["bind_addresses"], + listener["port"], + manhole( + username="matrix", + password="rabbithole", + globals={"hs": self}, ) + ) else: logger.warn("Unrecognized listener type: %s", listener["type"]) diff --git a/synapse/app/synctl.py b/synapse/app/synctl.py index 3bd7ef7bba..0f0ddfa78a 100755 --- a/synapse/app/synctl.py +++ b/synapse/app/synctl.py @@ -184,6 +184,9 @@ def main(): worker_configfiles.append(worker_configfile) if options.all_processes: + # To start the main synapse with -a you need to add a worker file + # with worker_app == "synapse.app.homeserver" + start_stop_synapse = False worker_configdir = options.all_processes if not os.path.isdir(worker_configdir): write( @@ -200,11 +203,29 @@ def main(): with open(worker_configfile) as stream: worker_config = yaml.load(stream) worker_app = worker_config["worker_app"] - worker_pidfile = worker_config["worker_pid_file"] - worker_daemonize = worker_config["worker_daemonize"] - assert worker_daemonize, "In config %r: expected '%s' to be True" % ( - worker_configfile, "worker_daemonize") - worker_cache_factor = worker_config.get("synctl_cache_factor") + if worker_app == "synapse.app.homeserver": + # We need to special case all of this to pick up options that may + # be set in the main config file or in this worker config file. + worker_pidfile = ( + worker_config.get("pid_file") + or pidfile + ) + worker_cache_factor = worker_config.get("synctl_cache_factor") or cache_factor + daemonize = worker_config.get("daemonize") or config.get("daemonize") + assert daemonize, "Main process must have daemonize set to true" + + # The master process doesn't support using worker_* config. + for key in worker_config: + if key == "worker_app": # But we allow worker_app + continue + assert not key.startswith("worker_"), \ + "Main process cannot use worker_* config" + else: + worker_pidfile = worker_config["worker_pid_file"] + worker_daemonize = worker_config["worker_daemonize"] + assert worker_daemonize, "In config %r: expected '%s' to be True" % ( + worker_configfile, "worker_daemonize") + worker_cache_factor = worker_config.get("synctl_cache_factor") workers.append(Worker( worker_app, worker_configfile, worker_pidfile, worker_cache_factor, )) diff --git a/synapse/app/user_dir.py b/synapse/app/user_dir.py index be661a70c7..494ccb702c 100644 --- a/synapse/app/user_dir.py +++ b/synapse/app/user_dir.py @@ -92,19 +92,6 @@ class UserDirectorySlaveStore( class UserDirectoryServer(HomeServer): - def get_db_conn(self, run_new_connection=True): - # Any param beginning with cp_ is a parameter for adbapi, and should - # not be passed to the database engine. - db_params = { - k: v for k, v in self.db_config.get("args", {}).items() - if not k.startswith("cp_") - } - db_conn = self.database_engine.module.connect(**db_params) - - if run_new_connection: - self.database_engine.on_new_connection(db_conn) - return db_conn - def setup(self): logger.info("Setting up.") self.datastore = UserDirectorySlaveStore(self.get_db_conn(), self) @@ -131,17 +118,16 @@ class UserDirectoryServer(HomeServer): root_resource = create_resource_tree(resources, Resource()) - for address in bind_addresses: - reactor.listenTCP( - port, - SynapseSite( - "synapse.access.http.%s" % (site_tag,), - site_tag, - listener_config, - root_resource, - ), - interface=address + _base.listen_tcp( + bind_addresses, + port, + SynapseSite( + "synapse.access.http.%s" % (site_tag,), + site_tag, + listener_config, + root_resource, ) + ) logger.info("Synapse user_dir now listening on port %d", port) @@ -150,18 +136,15 @@ class UserDirectoryServer(HomeServer): if listener["type"] == "http": self._listen_http(listener) elif listener["type"] == "manhole": - bind_addresses = listener["bind_addresses"] - - for address in bind_addresses: - reactor.listenTCP( - listener["port"], - manhole( - username="matrix", - password="rabbithole", - globals={"hs": self}, - ), - interface=address + _base.listen_tcp( + listener["bind_addresses"], + listener["port"], + manhole( + username="matrix", + password="rabbithole", + globals={"hs": self}, ) + ) else: logger.warn("Unrecognized listener type: %s", listener["type"]) diff --git a/synapse/config/logger.py b/synapse/config/logger.py index a1d6e4d4f7..3f70039acd 100644 --- a/synapse/config/logger.py +++ b/synapse/config/logger.py @@ -28,27 +28,27 @@ DEFAULT_LOG_CONFIG = Template(""" version: 1 formatters: - precise: - format: '%(asctime)s - %(name)s - %(lineno)d - %(levelname)s - %(request)s\ -- %(message)s' + precise: + format: '%(asctime)s - %(name)s - %(lineno)d - %(levelname)s - \ +%(request)s - %(message)s' filters: - context: - (): synapse.util.logcontext.LoggingContextFilter - request: "" + context: + (): synapse.util.logcontext.LoggingContextFilter + request: "" handlers: - file: - class: logging.handlers.RotatingFileHandler - formatter: precise - filename: ${log_file} - maxBytes: 104857600 - backupCount: 10 - filters: [context] - console: - class: logging.StreamHandler - formatter: precise - filters: [context] + file: + class: logging.handlers.RotatingFileHandler + formatter: precise + filename: ${log_file} + maxBytes: 104857600 + backupCount: 10 + filters: [context] + console: + class: logging.StreamHandler + formatter: precise + filters: [context] loggers: synapse: @@ -74,17 +74,10 @@ class LoggingConfig(Config): self.log_file = self.abspath(config.get("log_file")) def default_config(self, config_dir_path, server_name, **kwargs): - log_file = self.abspath("homeserver.log") log_config = self.abspath( os.path.join(config_dir_path, server_name + ".log.config") ) return """ - # Logging verbosity level. Ignored if log_config is specified. - verbose: 0 - - # File to write logging to. Ignored if log_config is specified. - log_file: "%(log_file)s" - # A yaml python logging config file log_config: "%(log_config)s" """ % locals() @@ -123,9 +116,10 @@ class LoggingConfig(Config): def generate_files(self, config): log_config = config.get("log_config") if log_config and not os.path.exists(log_config): + log_file = self.abspath("homeserver.log") with open(log_config, "wb") as log_config_file: log_config_file.write( - DEFAULT_LOG_CONFIG.substitute(log_file=config["log_file"]) + DEFAULT_LOG_CONFIG.substitute(log_file=log_file) ) @@ -150,6 +144,9 @@ def setup_logging(config, use_worker_options=False): ) if log_config is None: + # We don't have a logfile, so fall back to the 'verbosity' param from + # the config or cmdline. (Note that we generate a log config for new + # installs, so this will be an unusual case) level = logging.INFO level_for_storage = logging.INFO if config.verbosity: @@ -157,11 +154,10 @@ def setup_logging(config, use_worker_options=False): if config.verbosity > 1: level_for_storage = logging.DEBUG - # FIXME: we need a logging.WARN for a -q quiet option logger = logging.getLogger('') logger.setLevel(level) - logging.getLogger('synapse.storage').setLevel(level_for_storage) + logging.getLogger('synapse.storage.SQL').setLevel(level_for_storage) formatter = logging.Formatter(log_format) if log_file: diff --git a/synapse/config/password_auth_providers.py b/synapse/config/password_auth_providers.py index e9828fac17..6602c5b4c7 100644 --- a/synapse/config/password_auth_providers.py +++ b/synapse/config/password_auth_providers.py @@ -29,10 +29,10 @@ class PasswordAuthProviderConfig(Config): # param. ldap_config = config.get("ldap_config", {}) if ldap_config.get("enabled", False): - providers.append[{ + providers.append({ 'module': LDAP_PROVIDER, 'config': ldap_config, - }] + }) providers.extend(config.get("password_providers", [])) for provider in providers: diff --git a/synapse/config/registration.py b/synapse/config/registration.py index ef917fc9f2..c5384b3ad4 100644 --- a/synapse/config/registration.py +++ b/synapse/config/registration.py @@ -31,6 +31,8 @@ class RegistrationConfig(Config): strtobool(str(config["disable_registration"])) ) + self.registrations_require_3pid = config.get("registrations_require_3pid", []) + self.allowed_local_3pids = config.get("allowed_local_3pids", []) self.registration_shared_secret = config.get("registration_shared_secret") self.bcrypt_rounds = config.get("bcrypt_rounds", 12) @@ -52,13 +54,32 @@ class RegistrationConfig(Config): # Enable registration for new users. enable_registration: False + # The user must provide all of the below types of 3PID when registering. + # + # registrations_require_3pid: + # - email + # - msisdn + + # Mandate that users are only allowed to associate certain formats of + # 3PIDs with accounts on this server. + # + # allowed_local_3pids: + # - medium: email + # pattern: ".*@matrix\\.org" + # - medium: email + # pattern: ".*@vector\\.im" + # - medium: msisdn + # pattern: "\\+44" + # If set, allows registration by anyone who also has the shared # secret, even if registration is otherwise disabled. registration_shared_secret: "%(registration_shared_secret)s" # Set the number of bcrypt rounds used to generate password hash. # Larger numbers increase the work factor needed to generate the hash. - # The default number of rounds is 12. + # The default number is 12 (which equates to 2^12 rounds). + # N.B. that increasing this will exponentially increase the time required + # to register or login - e.g. 24 => 2^24 rounds which will take >20 mins. bcrypt_rounds: 12 # Allows users to register as guests without a password/email/etc, and diff --git a/synapse/config/repository.py b/synapse/config/repository.py index 6baa474931..25ea77738a 100644 --- a/synapse/config/repository.py +++ b/synapse/config/repository.py @@ -16,6 +16,8 @@ from ._base import Config, ConfigError from collections import namedtuple +from synapse.util.module_loader import load_module + MISSING_NETADDR = ( "Missing netaddr library. This is required for URL preview API." @@ -36,6 +38,14 @@ ThumbnailRequirement = namedtuple( "ThumbnailRequirement", ["width", "height", "method", "media_type"] ) +MediaStorageProviderConfig = namedtuple( + "MediaStorageProviderConfig", ( + "store_local", # Whether to store newly uploaded local files + "store_remote", # Whether to store newly downloaded remote files + "store_synchronous", # Whether to wait for successful storage for local uploads + ), +) + def parse_thumbnail_requirements(thumbnail_sizes): """ Takes a list of dictionaries with "width", "height", and "method" keys @@ -73,16 +83,61 @@ class ContentRepositoryConfig(Config): self.media_store_path = self.ensure_directory(config["media_store_path"]) - self.backup_media_store_path = config.get("backup_media_store_path") - if self.backup_media_store_path: - self.backup_media_store_path = self.ensure_directory( - self.backup_media_store_path - ) + backup_media_store_path = config.get("backup_media_store_path") - self.synchronous_backup_media_store = config.get( + synchronous_backup_media_store = config.get( "synchronous_backup_media_store", False ) + storage_providers = config.get("media_storage_providers", []) + + if backup_media_store_path: + if storage_providers: + raise ConfigError( + "Cannot use both 'backup_media_store_path' and 'storage_providers'" + ) + + storage_providers = [{ + "module": "file_system", + "store_local": True, + "store_synchronous": synchronous_backup_media_store, + "store_remote": True, + "config": { + "directory": backup_media_store_path, + } + }] + + # This is a list of config that can be used to create the storage + # providers. The entries are tuples of (Class, class_config, + # MediaStorageProviderConfig), where Class is the class of the provider, + # the class_config the config to pass to it, and + # MediaStorageProviderConfig are options for StorageProviderWrapper. + # + # We don't create the storage providers here as not all workers need + # them to be started. + self.media_storage_providers = [] + + for provider_config in storage_providers: + # We special case the module "file_system" so as not to need to + # expose FileStorageProviderBackend + if provider_config["module"] == "file_system": + provider_config["module"] = ( + "synapse.rest.media.v1.storage_provider" + ".FileStorageProviderBackend" + ) + + provider_class, parsed_config = load_module(provider_config) + + wrapper_config = MediaStorageProviderConfig( + provider_config.get("store_local", False), + provider_config.get("store_remote", False), + provider_config.get("store_synchronous", False), + ) + + self.media_storage_providers.append( + (provider_class, parsed_config, wrapper_config,) + ) + self.uploads_path = self.ensure_directory(config["uploads_path"]) self.dynamic_thumbnails = config["dynamic_thumbnails"] self.thumbnail_requirements = parse_thumbnail_requirements( @@ -127,13 +182,19 @@ class ContentRepositoryConfig(Config): # Directory where uploaded images and attachments are stored. media_store_path: "%(media_store)s" - # A secondary directory where uploaded images and attachments are - # stored as a backup. - # backup_media_store_path: "%(media_store)s" - - # Whether to wait for successful write to backup media store before - # returning successfully. - # synchronous_backup_media_store: false + # Media storage providers allow media to be stored in different + # locations. + # media_storage_providers: + # - module: file_system + # # Whether to write new local files. + # store_local: false + # # Whether to write new remote media + # store_remote: false + # # Whether to block upload requests waiting for write to this + # # provider to complete + # store_synchronous: false + # config: + # directory: /mnt/some/other/directory # Directory where in-progress uploads are stored. uploads_path: "%(uploads_path)s" diff --git a/synapse/config/server.py b/synapse/config/server.py index edb90a1348..8f0b6d1f28 100644 --- a/synapse/config/server.py +++ b/synapse/config/server.py @@ -55,6 +55,17 @@ class ServerConfig(Config): "block_non_admin_invites", False, ) + # FIXME: federation_domain_whitelist needs sytests + self.federation_domain_whitelist = None + federation_domain_whitelist = config.get( + "federation_domain_whitelist", None + ) + # turn the whitelist into a hash for speed of lookup + if federation_domain_whitelist is not None: + self.federation_domain_whitelist = {} + for domain in federation_domain_whitelist: + self.federation_domain_whitelist[domain] = True + if self.public_baseurl is not None: if self.public_baseurl[-1] != '/': self.public_baseurl += '/' @@ -210,6 +221,17 @@ class ServerConfig(Config): # (except those sent by local server admins). The default is False. # block_non_admin_invites: True + # Restrict federation to the following whitelist of domains. + # N.B. we recommend also firewalling your federation listener to limit + # inbound federation traffic as early as possible, rather than relying + # purely on this application-layer restriction. If not specified, the + # default is to whitelist everything. + # + # federation_domain_whitelist: + # - lon.example.com + # - nyc.example.com + # - syd.example.com + # List of ports that Synapse should listen on, their purpose and their # configuration. listeners: @@ -220,13 +242,12 @@ class ServerConfig(Config): port: %(bind_port)s # Local addresses to listen on. - # This will listen on all IPv4 addresses by default. + # On Linux and Mac OS, `::` will listen on all IPv4 and IPv6 + # addresses by default. For most other OSes, this will only listen + # on IPv6. bind_addresses: + - '::' - '0.0.0.0' - # Uncomment to listen on all IPv6 interfaces - # N.B: On at least Linux this will also listen on all IPv4 - # addresses, so you will need to comment out the line above. - # - '::' # This is a 'http' listener, allows us to specify 'resources'. type: http @@ -264,7 +285,7 @@ class ServerConfig(Config): # For when matrix traffic passes through loadbalancer that unwraps TLS. - port: %(unsecure_port)s tls: false - bind_addresses: ['0.0.0.0'] + bind_addresses: ['::', '0.0.0.0'] type: http x_forwarded: false @@ -278,7 +299,7 @@ class ServerConfig(Config): # Turn on the twisted ssh manhole service on localhost on the given # port. # - port: 9000 - # bind_address: 127.0.0.1 + # bind_addresses: ['::1', '127.0.0.1'] # type: manhole """ % locals() diff --git a/synapse/config/tls.py b/synapse/config/tls.py index 4748f71c2f..29eb012ddb 100644 --- a/synapse/config/tls.py +++ b/synapse/config/tls.py @@ -96,7 +96,7 @@ class TlsConfig(Config): # certificates returned by this server match one of the fingerprints. # # Synapse automatically adds the fingerprint of its own certificate - # to the list. So if federation traffic is handle directly by synapse + # to the list. So if federation traffic is handled directly by synapse # then no modification to the list is required. # # If synapse is run behind a load balancer that handles the TLS then it diff --git a/synapse/config/workers.py b/synapse/config/workers.py index c5a5a8919c..80baf0ce0e 100644 --- a/synapse/config/workers.py +++ b/synapse/config/workers.py @@ -23,13 +23,26 @@ class WorkerConfig(Config): def read_config(self, config): self.worker_app = config.get("worker_app") + + # Canonicalise worker_app so that master always has None + if self.worker_app == "synapse.app.homeserver": + self.worker_app = None + self.worker_listeners = config.get("worker_listeners") self.worker_daemonize = config.get("worker_daemonize") self.worker_pid_file = config.get("worker_pid_file") self.worker_log_file = config.get("worker_log_file") self.worker_log_config = config.get("worker_log_config") + + # The host used to connect to the main synapse self.worker_replication_host = config.get("worker_replication_host", None) + + # The port on the main synapse for TCP replication self.worker_replication_port = config.get("worker_replication_port", None) + + # The port on the main synapse for HTTP replication endpoint + self.worker_replication_http_port = config.get("worker_replication_http_port") + self.worker_name = config.get("worker_name", self.worker_app) self.worker_main_http_uri = config.get("worker_main_http_uri", None) diff --git a/synapse/event_auth.py b/synapse/event_auth.py index 061ee86b16..cd5627e36a 100644 --- a/synapse/event_auth.py +++ b/synapse/event_auth.py @@ -319,7 +319,7 @@ def _is_membership_change_allowed(event, auth_events): # TODO (erikj): Implement kicks. if target_banned and user_level < ban_level: raise AuthError( - 403, "You cannot unban user &s." % (target_user_id,) + 403, "You cannot unban user %s." % (target_user_id,) ) elif target_user_id != event.user_id: kick_level = _get_named_level(auth_events, "kick", 50) diff --git a/synapse/events/snapshot.py b/synapse/events/snapshot.py index e9a732ff03..8e684d91b5 100644 --- a/synapse/events/snapshot.py +++ b/synapse/events/snapshot.py @@ -13,6 +13,10 @@ # See the License for the specific language governing permissions and # limitations under the License. +from twisted.internet import defer + +from frozendict import frozendict + class EventContext(object): """ @@ -25,7 +29,9 @@ class EventContext(object): The current state map excluding the current event. (type, state_key) -> event_id - state_group (int): state group id + state_group (int|None): state group id, if the state has been stored + as a state group. This is usually only None if e.g. the event is + an outlier. rejected (bool|str): A rejection reason if the event was rejected, else False @@ -46,7 +52,6 @@ class EventContext(object): "prev_state_ids", "state_group", "rejected", - "push_actions", "prev_group", "delta_ids", "prev_state_events", @@ -61,7 +66,6 @@ class EventContext(object): self.state_group = None self.rejected = False - self.push_actions = [] # A previously persisted state group and a delta between that # and this state. @@ -71,3 +75,98 @@ class EventContext(object): self.prev_state_events = None self.app_service = None + + def serialize(self, event): + """Converts self to a type that can be serialized as JSON, and then + deserialized by `deserialize` + + Args: + event (FrozenEvent): The event that this context relates to + + Returns: + dict + """ + + # We don't serialize the full state dicts, instead they get pulled out + # of the DB on the other side. However, the other side can't figure out + # the prev_state_ids, so if we're a state event we include the event + # id that we replaced in the state. + if event.is_state(): + prev_state_id = self.prev_state_ids.get((event.type, event.state_key)) + else: + prev_state_id = None + + return { + "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, + "rejected": self.rejected, + "prev_group": self.prev_group, + "delta_ids": _encode_state_dict(self.delta_ids), + "prev_state_events": self.prev_state_events, + "app_service_id": self.app_service.id if self.app_service else None + } + + @staticmethod + @defer.inlineCallbacks + def deserialize(store, input): + """Converts a dict that was produced by `serialize` back into a + EventContext. + + Args: + store (DataStore): Used to convert AS ID to AS object + input (dict): A dict produced by `serialize` + + Returns: + EventContext + """ + context = EventContext() + context.state_group = input["state_group"] + context.rejected = input["rejected"] + context.prev_group = input["prev_group"] + context.delta_ids = _decode_state_dict(input["delta_ids"]) + context.prev_state_events = input["prev_state_events"] + + # We use the state_group and prev_state_id stuff to pull the + # current_state_ids out of the DB and construct prev_state_ids. + prev_state_id = input["prev_state_id"] + event_type = input["event_type"] + event_state_key = input["event_state_key"] + + context.current_state_ids = yield store.get_state_ids_for_group( + context.state_group, + ) + if prev_state_id and event_state_key: + context.prev_state_ids = dict(context.current_state_ids) + context.prev_state_ids[(event_type, event_state_key)] = prev_state_id + else: + context.prev_state_ids = context.current_state_ids + + app_service_id = input["app_service_id"] + if app_service_id: + context.app_service = store.get_app_service_by_id(app_service_id) + + defer.returnValue(context) + + +def _encode_state_dict(state_dict): + """Since dicts of (type, state_key) -> event_id cannot be serialized in + JSON we need to convert them to a form that can. + """ + if state_dict is None: + return None + + return [ + (etype, state_key, v) + for (etype, state_key), v in state_dict.iteritems() + ] + + +def _decode_state_dict(input): + """Decodes a state dict encoded using `_encode_state_dict` above + """ + if input is None: + return None + + return frozendict({(etype, state_key,): v for etype, state_key, v in input}) diff --git a/synapse/federation/__init__.py b/synapse/federation/__init__.py index 2e32d245ba..f5f0bdfca3 100644 --- a/synapse/federation/__init__.py +++ b/synapse/federation/__init__.py @@ -15,11 +15,3 @@ """ This package includes all the federation specific logic. """ - -from .replication import ReplicationLayer - - -def initialize_http_replication(hs): - transport = hs.get_federation_transport_client() - - return ReplicationLayer(hs, transport) diff --git a/synapse/federation/federation_base.py b/synapse/federation/federation_base.py index a0f5d40eb3..79eaa31031 100644 --- a/synapse/federation/federation_base.py +++ b/synapse/federation/federation_base.py @@ -16,7 +16,9 @@ import logging from synapse.api.errors import SynapseError from synapse.crypto.event_signing import check_event_content_hash +from synapse.events import FrozenEvent from synapse.events.utils import prune_event +from synapse.http.servlet import assert_params_in_request from synapse.util import unwrapFirstError, logcontext from twisted.internet import defer @@ -25,7 +27,13 @@ logger = logging.getLogger(__name__) class FederationBase(object): def __init__(self, hs): + self.hs = hs + + self.server_name = hs.hostname + self.keyring = hs.get_keyring() self.spam_checker = hs.get_spam_checker() + self.store = hs.get_datastore() + self._clock = hs.get_clock() @defer.inlineCallbacks def _check_sigs_and_hash_and_fetch(self, origin, pdus, outlier=False, @@ -169,3 +177,28 @@ class FederationBase(object): ) return deferreds + + +def event_from_pdu_json(pdu_json, outlier=False): + """Construct a FrozenEvent from an event json received over federation + + Args: + pdu_json (object): pdu as received over federation + outlier (bool): True to mark this event as an outlier + + Returns: + FrozenEvent + + Raises: + SynapseError: if the pdu is missing required fields + """ + # we could probably enforce a bunch of other fields here (room_id, sender, + # origin, etc etc) + assert_params_in_request(pdu_json, ('event_id', 'type')) + event = FrozenEvent( + pdu_json + ) + + event.internal_metadata.outlier = outlier + + return event diff --git a/synapse/federation/federation_client.py b/synapse/federation/federation_client.py index b8f02f5391..38440da5b5 100644 --- a/synapse/federation/federation_client.py +++ b/synapse/federation/federation_client.py @@ -14,29 +14,29 @@ # limitations under the License. +import copy +import itertools +import logging +import random + from twisted.internet import defer -from .federation_base import FederationBase from synapse.api.constants import Membership - from synapse.api.errors import ( - CodeMessageException, HttpResponseException, SynapseError, + CodeMessageException, HttpResponseException, SynapseError, FederationDeniedError ) -from synapse.util import unwrapFirstError, logcontext +from synapse.events import builder +from synapse.federation.federation_base import ( + FederationBase, + event_from_pdu_json, +) +import synapse.metrics +from synapse.util import logcontext, unwrapFirstError from synapse.util.caches.expiringcache import ExpiringCache -from synapse.util.logutils import log_function from synapse.util.logcontext import make_deferred_yieldable, preserve_fn -from synapse.events import FrozenEvent, builder -import synapse.metrics - +from synapse.util.logutils import log_function from synapse.util.retryutils import NotRetryingDestination -import copy -import itertools -import logging -import random - - logger = logging.getLogger(__name__) @@ -58,6 +58,7 @@ class FederationClient(FederationBase): self._clear_tried_cache, 60 * 1000, ) self.state = hs.get_state_handler() + self.transport_layer = hs.get_federation_transport_client() def _clear_tried_cache(self): """Clear pdu_destination_tried cache""" @@ -184,7 +185,7 @@ class FederationClient(FederationBase): logger.debug("backfill transaction_data=%s", repr(transaction_data)) pdus = [ - self.event_from_pdu_json(p, outlier=False) + event_from_pdu_json(p, outlier=False) for p in transaction_data["pdus"] ] @@ -244,7 +245,7 @@ class FederationClient(FederationBase): logger.debug("transaction_data %r", transaction_data) pdu_list = [ - self.event_from_pdu_json(p, outlier=outlier) + event_from_pdu_json(p, outlier=outlier) for p in transaction_data["pdus"] ] @@ -266,6 +267,9 @@ class FederationClient(FederationBase): except NotRetryingDestination as e: logger.info(e.message) continue + except FederationDeniedError as e: + logger.info(e.message) + continue except Exception as e: pdu_attempts[destination] = now @@ -336,11 +340,11 @@ class FederationClient(FederationBase): ) pdus = [ - self.event_from_pdu_json(p, outlier=True) for p in result["pdus"] + event_from_pdu_json(p, outlier=True) for p in result["pdus"] ] auth_chain = [ - self.event_from_pdu_json(p, outlier=True) + event_from_pdu_json(p, outlier=True) for p in result.get("auth_chain", []) ] @@ -441,7 +445,7 @@ class FederationClient(FederationBase): ) auth_chain = [ - self.event_from_pdu_json(p, outlier=True) + event_from_pdu_json(p, outlier=True) for p in res["auth_chain"] ] @@ -570,12 +574,12 @@ class FederationClient(FederationBase): logger.debug("Got content: %s", content) state = [ - self.event_from_pdu_json(p, outlier=True) + event_from_pdu_json(p, outlier=True) for p in content.get("state", []) ] auth_chain = [ - self.event_from_pdu_json(p, outlier=True) + event_from_pdu_json(p, outlier=True) for p in content.get("auth_chain", []) ] @@ -650,7 +654,7 @@ class FederationClient(FederationBase): logger.debug("Got response to send_invite: %s", pdu_dict) - pdu = self.event_from_pdu_json(pdu_dict) + pdu = event_from_pdu_json(pdu_dict) # Check signatures are correct. pdu = yield self._check_sigs_and_hash(pdu) @@ -740,7 +744,7 @@ class FederationClient(FederationBase): ) auth_chain = [ - self.event_from_pdu_json(e) + event_from_pdu_json(e) for e in content["auth_chain"] ] @@ -788,7 +792,7 @@ class FederationClient(FederationBase): ) events = [ - self.event_from_pdu_json(e) + event_from_pdu_json(e) for e in content.get("events", []) ] @@ -805,15 +809,6 @@ class FederationClient(FederationBase): defer.returnValue(signed_events) - def event_from_pdu_json(self, pdu_json, outlier=False): - event = FrozenEvent( - pdu_json - ) - - event.internal_metadata.outlier = outlier - - return event - @defer.inlineCallbacks def forward_third_party_invite(self, destinations, room_id, event_dict): for destination in destinations: diff --git a/synapse/federation/federation_server.py b/synapse/federation/federation_server.py index a2327f24b6..bea7fd0b71 100644 --- a/synapse/federation/federation_server.py +++ b/synapse/federation/federation_server.py @@ -12,25 +12,26 @@ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. +import logging + +import simplejson as json from twisted.internet import defer -from .federation_base import FederationBase -from .units import Transaction, Edu +from synapse.api.errors import AuthError, FederationError, SynapseError, NotFoundError +from synapse.crypto.event_signing import compute_event_signature +from synapse.federation.federation_base import ( + FederationBase, + event_from_pdu_json, +) +from synapse.federation.persistence import TransactionActions +from synapse.federation.units import Edu, Transaction +import synapse.metrics +from synapse.types import get_domain_from_id from synapse.util import async +from synapse.util.caches.response_cache import ResponseCache from synapse.util.logcontext import make_deferred_yieldable, preserve_fn from synapse.util.logutils import log_function -from synapse.util.caches.response_cache import ResponseCache -from synapse.events import FrozenEvent -from synapse.types import get_domain_from_id -import synapse.metrics - -from synapse.api.errors import AuthError, FederationError, SynapseError - -from synapse.crypto.event_signing import compute_event_signature - -import simplejson as json -import logging # when processing incoming transactions, we try to handle multiple rooms in # parallel, up to this limit. @@ -53,50 +54,19 @@ class FederationServer(FederationBase): super(FederationServer, self).__init__(hs) self.auth = hs.get_auth() + self.handler = hs.get_handlers().federation_handler self._server_linearizer = async.Linearizer("fed_server") self._transaction_linearizer = async.Linearizer("fed_txn_handler") + self.transaction_actions = TransactionActions(self.store) + + self.registry = hs.get_federation_registry() + # We cache responses to state queries, as they take a while and often # come in waves. self._state_resp_cache = ResponseCache(hs, timeout_ms=30000) - def set_handler(self, handler): - """Sets the handler that the replication layer will use to communicate - receipt of new PDUs from other home servers. The required methods are - documented on :py:class:`.ReplicationHandler`. - """ - self.handler = handler - - def register_edu_handler(self, edu_type, handler): - if edu_type in self.edu_handlers: - raise KeyError("Already have an EDU handler for %s" % (edu_type,)) - - self.edu_handlers[edu_type] = handler - - def register_query_handler(self, query_type, handler): - """Sets the handler callable that will be used to handle an incoming - federation Query of the given type. - - Args: - query_type (str): Category name of the query, which should match - the string used by make_query. - handler (callable): Invoked to handle incoming queries of this type - - handler is invoked as: - result = handler(args) - - where 'args' is a dict mapping strings to strings of the query - arguments. It should return a Deferred that will eventually yield an - object to encode as JSON. - """ - if query_type in self.query_handlers: - raise KeyError( - "Already have a Query handler for %s" % (query_type,) - ) - - self.query_handlers[query_type] = handler - @defer.inlineCallbacks @log_function def on_backfill_request(self, origin, room_id, versions, limit): @@ -172,7 +142,7 @@ class FederationServer(FederationBase): p["age_ts"] = request_time - int(p["age"]) del p["age"] - event = self.event_from_pdu_json(p) + event = event_from_pdu_json(p) room_id = event.room_id pdus_by_room.setdefault(room_id, []).append(event) @@ -230,16 +200,7 @@ class FederationServer(FederationBase): @defer.inlineCallbacks def received_edu(self, origin, edu_type, content): received_edus_counter.inc() - - if edu_type in self.edu_handlers: - try: - yield self.edu_handlers[edu_type](origin, content) - except SynapseError as e: - logger.info("Failed to handle edu %r: %r", edu_type, e) - except Exception as e: - logger.exception("Failed to handle edu %r", edu_type) - else: - logger.warn("Received EDU of type %s with no handler", edu_type) + yield self.registry.on_edu(edu_type, origin, content) @defer.inlineCallbacks @log_function @@ -329,14 +290,8 @@ class FederationServer(FederationBase): @defer.inlineCallbacks def on_query_request(self, query_type, args): received_queries_counter.inc(query_type) - - if query_type in self.query_handlers: - response = yield self.query_handlers[query_type](args) - defer.returnValue((200, response)) - else: - defer.returnValue( - (404, "No handler for Query type '%s'" % (query_type,)) - ) + resp = yield self.registry.on_query(query_type, args) + defer.returnValue((200, resp)) @defer.inlineCallbacks def on_make_join_request(self, room_id, user_id): @@ -346,7 +301,7 @@ class FederationServer(FederationBase): @defer.inlineCallbacks def on_invite_request(self, origin, content): - pdu = self.event_from_pdu_json(content) + pdu = event_from_pdu_json(content) ret_pdu = yield self.handler.on_invite_request(origin, pdu) time_now = self._clock.time_msec() defer.returnValue((200, {"event": ret_pdu.get_pdu_json(time_now)})) @@ -354,7 +309,7 @@ class FederationServer(FederationBase): @defer.inlineCallbacks def on_send_join_request(self, origin, content): logger.debug("on_send_join_request: content: %s", content) - pdu = self.event_from_pdu_json(content) + pdu = event_from_pdu_json(content) logger.debug("on_send_join_request: pdu sigs: %s", pdu.signatures) res_pdus = yield self.handler.on_send_join_request(origin, pdu) time_now = self._clock.time_msec() @@ -374,7 +329,7 @@ class FederationServer(FederationBase): @defer.inlineCallbacks def on_send_leave_request(self, origin, content): logger.debug("on_send_leave_request: content: %s", content) - pdu = self.event_from_pdu_json(content) + pdu = event_from_pdu_json(content) logger.debug("on_send_leave_request: pdu sigs: %s", pdu.signatures) yield self.handler.on_send_leave_request(origin, pdu) defer.returnValue((200, {})) @@ -411,7 +366,7 @@ class FederationServer(FederationBase): """ with (yield self._server_linearizer.queue((origin, room_id))): auth_chain = [ - self.event_from_pdu_json(e) + event_from_pdu_json(e) for e in content["auth_chain"] ] @@ -586,15 +541,6 @@ class FederationServer(FederationBase): def __str__(self): return "<ReplicationLayer(%s)>" % self.server_name - def event_from_pdu_json(self, pdu_json, outlier=False): - event = FrozenEvent( - pdu_json - ) - - event.internal_metadata.outlier = outlier - - return event - @defer.inlineCallbacks def exchange_third_party_invite( self, @@ -617,3 +563,66 @@ class FederationServer(FederationBase): origin, room_id, event_dict ) defer.returnValue(ret) + + +class FederationHandlerRegistry(object): + """Allows classes to register themselves as handlers for a given EDU or + query type for incoming federation traffic. + """ + def __init__(self): + self.edu_handlers = {} + self.query_handlers = {} + + def register_edu_handler(self, edu_type, handler): + """Sets the handler callable that will be used to handle an incoming + federation EDU of the given type. + + Args: + edu_type (str): The type of the incoming EDU to register handler for + handler (Callable[[str, dict]]): A callable invoked on incoming EDU + of the given type. The arguments are the origin server name and + the EDU contents. + """ + if edu_type in self.edu_handlers: + raise KeyError("Already have an EDU handler for %s" % (edu_type,)) + + self.edu_handlers[edu_type] = handler + + def register_query_handler(self, query_type, handler): + """Sets the handler callable that will be used to handle an incoming + federation query of the given type. + + Args: + query_type (str): Category name of the query, which should match + the string used by make_query. + handler (Callable[[dict], Deferred[dict]]): Invoked to handle + incoming queries of this type. The return will be yielded + on and the result used as the response to the query request. + """ + if query_type in self.query_handlers: + raise KeyError( + "Already have a Query handler for %s" % (query_type,) + ) + + self.query_handlers[query_type] = handler + + @defer.inlineCallbacks + def on_edu(self, edu_type, origin, content): + handler = self.edu_handlers.get(edu_type) + if not handler: + logger.warn("No handler registered for EDU type %s", edu_type) + + try: + yield handler(origin, content) + except SynapseError as e: + logger.info("Failed to handle edu %r: %r", edu_type, e) + except Exception as e: + logger.exception("Failed to handle edu %r", edu_type) + + def on_query(self, query_type, args): + handler = self.query_handlers.get(query_type) + if not handler: + logger.warn("No handler registered for query type %s", query_type) + raise NotFoundError("No handler for Query type '%s'" % (query_type,)) + + return handler(args) diff --git a/synapse/federation/replication.py b/synapse/federation/replication.py deleted file mode 100644 index 62d865ec4b..0000000000 --- a/synapse/federation/replication.py +++ /dev/null @@ -1,73 +0,0 @@ -# -*- coding: utf-8 -*- -# Copyright 2014-2016 OpenMarket Ltd -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -"""This layer is responsible for replicating with remote home servers using -a given transport. -""" - -from .federation_client import FederationClient -from .federation_server import FederationServer - -from .persistence import TransactionActions - -import logging - - -logger = logging.getLogger(__name__) - - -class ReplicationLayer(FederationClient, FederationServer): - """This layer is responsible for replicating with remote home servers over - the given transport. I.e., does the sending and receiving of PDUs to - remote home servers. - - The layer communicates with the rest of the server via a registered - ReplicationHandler. - - In more detail, the layer: - * Receives incoming data and processes it into transactions and pdus. - * Fetches any PDUs it thinks it might have missed. - * Keeps the current state for contexts up to date by applying the - suitable conflict resolution. - * Sends outgoing pdus wrapped in transactions. - * Fills out the references to previous pdus/transactions appropriately - for outgoing data. - """ - - def __init__(self, hs, transport_layer): - self.server_name = hs.hostname - - self.keyring = hs.get_keyring() - - self.transport_layer = transport_layer - - self.federation_client = self - - self.store = hs.get_datastore() - - self.handler = None - self.edu_handlers = {} - self.query_handlers = {} - - self._clock = hs.get_clock() - - self.transaction_actions = TransactionActions(self.store) - - self.hs = hs - - super(ReplicationLayer, self).__init__(hs) - - def __str__(self): - return "<ReplicationLayer(%s)>" % self.server_name diff --git a/synapse/federation/transaction_queue.py b/synapse/federation/transaction_queue.py index 3e7809b04f..a141ec9953 100644 --- a/synapse/federation/transaction_queue.py +++ b/synapse/federation/transaction_queue.py @@ -19,7 +19,7 @@ from twisted.internet import defer from .persistence import TransactionActions from .units import Transaction, Edu -from synapse.api.errors import HttpResponseException +from synapse.api.errors import HttpResponseException, FederationDeniedError from synapse.util import logcontext, PreserveLoggingContext from synapse.util.async import run_on_reactor from synapse.util.retryutils import NotRetryingDestination, get_retry_limiter @@ -42,6 +42,8 @@ sent_edus_counter = client_metrics.register_counter("sent_edus") sent_transactions_counter = client_metrics.register_counter("sent_transactions") +events_processed_counter = client_metrics.register_counter("events_processed") + class TransactionQueue(object): """This class makes sure we only have one transaction in flight at @@ -205,6 +207,8 @@ class TransactionQueue(object): self._send_pdu(event, destinations) + events_processed_counter.inc_by(len(events)) + yield self.store.update_federation_out_pos( "events", next_token ) @@ -486,6 +490,8 @@ class TransactionQueue(object): (e.retry_last_ts + e.retry_interval) / 1000.0 ), ) + except FederationDeniedError as e: + logger.info(e) except Exception as e: logger.warn( "TX [%s] Failed to send transaction: %s", diff --git a/synapse/federation/transport/client.py b/synapse/federation/transport/client.py index 1f3ce238f6..5488e82985 100644 --- a/synapse/federation/transport/client.py +++ b/synapse/federation/transport/client.py @@ -212,6 +212,9 @@ class TransportLayerClient(object): Fails with ``NotRetryingDestination`` if we are not yet ready to retry this server. + + Fails with ``FederationDeniedError`` if the remote destination + is not in our federation whitelist """ valid_memberships = {Membership.JOIN, Membership.LEAVE} if membership not in valid_memberships: diff --git a/synapse/federation/transport/server.py b/synapse/federation/transport/server.py index 2b02b021ec..a66a6b0692 100644 --- a/synapse/federation/transport/server.py +++ b/synapse/federation/transport/server.py @@ -16,7 +16,7 @@ from twisted.internet import defer from synapse.api.urls import FEDERATION_PREFIX as PREFIX -from synapse.api.errors import Codes, SynapseError +from synapse.api.errors import Codes, SynapseError, FederationDeniedError from synapse.http.server import JsonResource from synapse.http.servlet import ( parse_json_object_from_request, parse_integer_from_args, parse_string_from_args, @@ -81,6 +81,7 @@ class Authenticator(object): self.keyring = hs.get_keyring() self.server_name = hs.hostname self.store = hs.get_datastore() + self.federation_domain_whitelist = hs.config.federation_domain_whitelist # A method just so we can pass 'self' as the authenticator to the Servlets @defer.inlineCallbacks @@ -92,6 +93,12 @@ class Authenticator(object): "signatures": {}, } + if ( + self.federation_domain_whitelist is not None and + self.server_name not in self.federation_domain_whitelist + ): + raise FederationDeniedError(self.server_name) + if content is not None: json_request["content"] = content @@ -1183,7 +1190,7 @@ GROUP_ATTESTATION_SERVLET_CLASSES = ( def register_servlets(hs, resource, authenticator, ratelimiter): for servletclass in FEDERATION_SERVLET_CLASSES: servletclass( - handler=hs.get_replication_layer(), + handler=hs.get_federation_server(), authenticator=authenticator, ratelimiter=ratelimiter, server_name=hs.hostname, diff --git a/synapse/handlers/__init__.py b/synapse/handlers/__init__.py index 53213cdccf..8f8fd82eb0 100644 --- a/synapse/handlers/__init__.py +++ b/synapse/handlers/__init__.py @@ -17,7 +17,6 @@ from .register import RegistrationHandler from .room import ( RoomCreationHandler, RoomContextHandler, ) -from .room_member import RoomMemberHandler from .message import MessageHandler from .federation import FederationHandler from .directory import DirectoryHandler @@ -49,7 +48,6 @@ class Handlers(object): self.registration_handler = RegistrationHandler(hs) self.message_handler = MessageHandler(hs) self.room_creation_handler = RoomCreationHandler(hs) - self.room_member_handler = RoomMemberHandler(hs) self.federation_handler = FederationHandler(hs) self.directory_handler = DirectoryHandler(hs) self.admin_handler = AdminHandler(hs) diff --git a/synapse/handlers/_base.py b/synapse/handlers/_base.py index faa5609c0c..e089e66fde 100644 --- a/synapse/handlers/_base.py +++ b/synapse/handlers/_base.py @@ -158,7 +158,7 @@ class BaseHandler(object): # homeserver. requester = synapse.types.create_requester( target_user, is_guest=True) - handler = self.hs.get_handlers().room_member_handler + handler = self.hs.get_room_member_handler() yield handler.update_membership( requester, target_user, diff --git a/synapse/handlers/appservice.py b/synapse/handlers/appservice.py index feca3e4c10..3dd3fa2a27 100644 --- a/synapse/handlers/appservice.py +++ b/synapse/handlers/appservice.py @@ -15,6 +15,7 @@ from twisted.internet import defer +import synapse from synapse.api.constants import EventTypes from synapse.util.metrics import Measure from synapse.util.logcontext import make_deferred_yieldable, preserve_fn @@ -23,6 +24,10 @@ import logging logger = logging.getLogger(__name__) +metrics = synapse.metrics.get_metrics_for(__name__) + +events_processed_counter = metrics.register_counter("events_processed") + def log_failure(failure): logger.error( @@ -103,6 +108,8 @@ class ApplicationServicesHandler(object): service, event ) + events_processed_counter.inc_by(len(events)) + yield self.store.set_appservice_last_pos(upper_bound) finally: self.is_processing = False diff --git a/synapse/handlers/auth.py b/synapse/handlers/auth.py index 573c9db8a1..a5365c4fe4 100644 --- a/synapse/handlers/auth.py +++ b/synapse/handlers/auth.py @@ -13,7 +13,7 @@ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. -from twisted.internet import defer +from twisted.internet import defer, threads from ._base import BaseHandler from synapse.api.constants import LoginType @@ -25,6 +25,7 @@ from synapse.module_api import ModuleApi from synapse.types import UserID from synapse.util.async import run_on_reactor from synapse.util.caches.expiringcache import ExpiringCache +from synapse.util.logcontext import make_deferred_yieldable from twisted.web.client import PartialDownloadError @@ -714,7 +715,7 @@ class AuthHandler(BaseHandler): if not lookupres: defer.returnValue(None) (user_id, password_hash) = lookupres - result = self.validate_hash(password, password_hash) + result = yield self.validate_hash(password, password_hash) if not result: logger.warn("Failed password login for user %s", user_id) defer.returnValue(None) @@ -842,10 +843,13 @@ class AuthHandler(BaseHandler): password (str): Password to hash. Returns: - Hashed password (str). + Deferred(str): Hashed password. """ - return bcrypt.hashpw(password.encode('utf8') + self.hs.config.password_pepper, - bcrypt.gensalt(self.bcrypt_rounds)) + def _do_hash(): + return bcrypt.hashpw(password.encode('utf8') + self.hs.config.password_pepper, + bcrypt.gensalt(self.bcrypt_rounds)) + + return make_deferred_yieldable(threads.deferToThread(_do_hash)) def validate_hash(self, password, stored_hash): """Validates that self.hash(password) == stored_hash. @@ -855,13 +859,19 @@ class AuthHandler(BaseHandler): stored_hash (str): Expected hash value. Returns: - Whether self.hash(password) == stored_hash (bool). + Deferred(bool): Whether self.hash(password) == stored_hash. """ + + def _do_validate_hash(): + return bcrypt.checkpw( + password.encode('utf8') + self.hs.config.password_pepper, + stored_hash.encode('utf8') + ) + if stored_hash: - return bcrypt.hashpw(password.encode('utf8') + self.hs.config.password_pepper, - stored_hash.encode('utf8')) == stored_hash + return make_deferred_yieldable(threads.deferToThread(_do_validate_hash)) else: - return False + return defer.succeed(False) class MacaroonGeneartor(object): diff --git a/synapse/handlers/device.py b/synapse/handlers/device.py index 2152efc692..40f3d24678 100644 --- a/synapse/handlers/device.py +++ b/synapse/handlers/device.py @@ -14,6 +14,7 @@ # limitations under the License. from synapse.api import errors from synapse.api.constants import EventTypes +from synapse.api.errors import FederationDeniedError from synapse.util import stringutils from synapse.util.async import Linearizer from synapse.util.caches.expiringcache import ExpiringCache @@ -36,14 +37,15 @@ class DeviceHandler(BaseHandler): self.state = hs.get_state_handler() self._auth_handler = hs.get_auth_handler() self.federation_sender = hs.get_federation_sender() - self.federation = hs.get_replication_layer() self._edu_updater = DeviceListEduUpdater(hs, self) - self.federation.register_edu_handler( + federation_registry = hs.get_federation_registry() + + federation_registry.register_edu_handler( "m.device_list_update", self._edu_updater.incoming_device_list_update, ) - self.federation.register_query_handler( + federation_registry.register_query_handler( "user_devices", self.on_federation_query_user_devices, ) @@ -429,7 +431,7 @@ class DeviceListEduUpdater(object): def __init__(self, hs, device_handler): self.store = hs.get_datastore() - self.federation = hs.get_replication_layer() + self.federation = hs.get_federation_client() self.clock = hs.get_clock() self.device_handler = device_handler @@ -513,6 +515,9 @@ class DeviceListEduUpdater(object): # This makes it more likely that the device lists will # eventually become consistent. return + except FederationDeniedError as e: + logger.info(e) + return except Exception: # TODO: Remember that we are now out of sync and try again # later diff --git a/synapse/handlers/devicemessage.py b/synapse/handlers/devicemessage.py index f7fad15c62..f147a20b73 100644 --- a/synapse/handlers/devicemessage.py +++ b/synapse/handlers/devicemessage.py @@ -17,7 +17,8 @@ import logging from twisted.internet import defer -from synapse.types import get_domain_from_id +from synapse.api.errors import SynapseError +from synapse.types import get_domain_from_id, UserID from synapse.util.stringutils import random_string @@ -33,10 +34,10 @@ class DeviceMessageHandler(object): """ self.store = hs.get_datastore() self.notifier = hs.get_notifier() - self.is_mine_id = hs.is_mine_id + self.is_mine = hs.is_mine self.federation = hs.get_federation_sender() - hs.get_replication_layer().register_edu_handler( + hs.get_federation_registry().register_edu_handler( "m.direct_to_device", self.on_direct_to_device_edu ) @@ -52,6 +53,12 @@ class DeviceMessageHandler(object): message_type = content["type"] message_id = content["message_id"] for user_id, by_device in content["messages"].items(): + # we use UserID.from_string to catch invalid user ids + if not self.is_mine(UserID.from_string(user_id)): + logger.warning("Request for keys for non-local user %s", + user_id) + raise SynapseError(400, "Not a user here") + messages_by_device = { device_id: { "content": message_content, @@ -77,7 +84,8 @@ class DeviceMessageHandler(object): local_messages = {} remote_messages = {} for user_id, by_device in messages.items(): - if self.is_mine_id(user_id): + # we use UserID.from_string to catch invalid user ids + if self.is_mine(UserID.from_string(user_id)): messages_by_device = { device_id: { "content": message_content, diff --git a/synapse/handlers/directory.py b/synapse/handlers/directory.py index a0464ae5c0..c5b6e75e03 100644 --- a/synapse/handlers/directory.py +++ b/synapse/handlers/directory.py @@ -34,9 +34,10 @@ class DirectoryHandler(BaseHandler): self.state = hs.get_state_handler() self.appservice_handler = hs.get_application_service_handler() + self.event_creation_handler = hs.get_event_creation_handler() - self.federation = hs.get_replication_layer() - self.federation.register_query_handler( + self.federation = hs.get_federation_client() + hs.get_federation_registry().register_query_handler( "directory", self.on_directory_query ) @@ -249,8 +250,7 @@ class DirectoryHandler(BaseHandler): def send_room_alias_update_event(self, requester, user_id, room_id): aliases = yield self.store.get_aliases_for_room(room_id) - msg_handler = self.hs.get_handlers().message_handler - yield msg_handler.create_and_send_nonmember_event( + yield self.event_creation_handler.create_and_send_nonmember_event( requester, { "type": EventTypes.Aliases, @@ -272,8 +272,7 @@ class DirectoryHandler(BaseHandler): if not alias_event or alias_event.content.get("alias", "") != alias_str: return - msg_handler = self.hs.get_handlers().message_handler - yield msg_handler.create_and_send_nonmember_event( + yield self.event_creation_handler.create_and_send_nonmember_event( requester, { "type": EventTypes.CanonicalAlias, diff --git a/synapse/handlers/e2e_keys.py b/synapse/handlers/e2e_keys.py index ce2c87e400..80b359b2e7 100644 --- a/synapse/handlers/e2e_keys.py +++ b/synapse/handlers/e2e_keys.py @@ -19,8 +19,10 @@ import logging from canonicaljson import encode_canonical_json from twisted.internet import defer -from synapse.api.errors import SynapseError, CodeMessageException -from synapse.types import get_domain_from_id +from synapse.api.errors import ( + SynapseError, CodeMessageException, FederationDeniedError, +) +from synapse.types import get_domain_from_id, UserID from synapse.util.logcontext import preserve_fn, make_deferred_yieldable from synapse.util.retryutils import NotRetryingDestination @@ -30,15 +32,15 @@ logger = logging.getLogger(__name__) class E2eKeysHandler(object): def __init__(self, hs): self.store = hs.get_datastore() - self.federation = hs.get_replication_layer() + self.federation = hs.get_federation_client() self.device_handler = hs.get_device_handler() - self.is_mine_id = hs.is_mine_id + self.is_mine = hs.is_mine self.clock = hs.get_clock() # doesn't really work as part of the generic query API, because the # query request requires an object POST, but we abuse the # "query handler" interface. - self.federation.register_query_handler( + hs.get_federation_registry().register_query_handler( "client_keys", self.on_federation_query_client_keys ) @@ -70,7 +72,8 @@ class E2eKeysHandler(object): remote_queries = {} for user_id, device_ids in device_keys_query.items(): - if self.is_mine_id(user_id): + # we use UserID.from_string to catch invalid user ids + if self.is_mine(UserID.from_string(user_id)): local_query[user_id] = device_ids else: remote_queries[user_id] = device_ids @@ -139,6 +142,10 @@ class E2eKeysHandler(object): failures[destination] = { "status": 503, "message": "Not ready for retry", } + except FederationDeniedError as e: + failures[destination] = { + "status": 403, "message": "Federation Denied", + } except Exception as e: # include ConnectionRefused and other errors failures[destination] = { @@ -170,7 +177,8 @@ class E2eKeysHandler(object): result_dict = {} for user_id, device_ids in query.items(): - if not self.is_mine_id(user_id): + # we use UserID.from_string to catch invalid user ids + if not self.is_mine(UserID.from_string(user_id)): logger.warning("Request for keys for non-local user %s", user_id) raise SynapseError(400, "Not a user here") @@ -213,7 +221,8 @@ class E2eKeysHandler(object): remote_queries = {} for user_id, device_keys in query.get("one_time_keys", {}).items(): - if self.is_mine_id(user_id): + # we use UserID.from_string to catch invalid user ids + if self.is_mine(UserID.from_string(user_id)): for device_id, algorithm in device_keys.items(): local_query.append((user_id, device_id, algorithm)) else: diff --git a/synapse/handlers/federation.py b/synapse/handlers/federation.py index ac70730885..080aca3d71 100644 --- a/synapse/handlers/federation.py +++ b/synapse/handlers/federation.py @@ -1,5 +1,6 @@ # -*- coding: utf-8 -*- # Copyright 2014-2016 OpenMarket Ltd +# Copyright 2018 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. @@ -22,6 +23,7 @@ from ._base import BaseHandler from synapse.api.errors import ( AuthError, FederationError, StoreError, CodeMessageException, SynapseError, + FederationDeniedError, ) from synapse.api.constants import EventTypes, Membership, RejectedReason from synapse.events.validator import EventValidator @@ -66,7 +68,7 @@ class FederationHandler(BaseHandler): self.hs = hs self.store = hs.get_datastore() - self.replication_layer = hs.get_replication_layer() + self.replication_layer = hs.get_federation_client() self.state_handler = hs.get_state_handler() self.server_name = hs.hostname self.keyring = hs.get_keyring() @@ -74,8 +76,7 @@ class FederationHandler(BaseHandler): self.is_mine_id = hs.is_mine_id self.pusher_pool = hs.get_pusherpool() self.spam_checker = hs.get_spam_checker() - - self.replication_layer.set_handler(self) + self.event_creation_handler = hs.get_event_creation_handler() # When joining a room we need to queue any events for that room up self.room_queues = {} @@ -782,6 +783,9 @@ class FederationHandler(BaseHandler): except NotRetryingDestination as e: logger.info(e.message) continue + except FederationDeniedError as e: + logger.info(e) + continue except Exception as e: logger.exception( "Failed to backfill from %s because %s", @@ -804,13 +808,12 @@ class FederationHandler(BaseHandler): event_ids = list(extremities.keys()) logger.debug("calling resolve_state_groups in _maybe_backfill") + resolve = logcontext.preserve_fn( + self.state_handler.resolve_state_groups_for_events + ) states = yield logcontext.make_deferred_yieldable(defer.gatherResults( - [ - logcontext.preserve_fn(self.state_handler.resolve_state_groups)( - room_id, [e] - ) - for e in event_ids - ], consumeErrors=True, + [resolve(room_id, [e]) for e in event_ids], + consumeErrors=True, )) states = dict(zip(event_ids, [s.state for s in states])) @@ -1004,8 +1007,7 @@ class FederationHandler(BaseHandler): }) try: - message_handler = self.hs.get_handlers().message_handler - event, context = yield message_handler._create_new_client_event( + event, context = yield self.event_creation_handler.create_new_client_event( builder=builder, ) except AuthError as e: @@ -1245,8 +1247,7 @@ class FederationHandler(BaseHandler): "state_key": user_id, }) - message_handler = self.hs.get_handlers().message_handler - event, context = yield message_handler._create_new_client_event( + event, context = yield self.event_creation_handler.create_new_client_event( builder=builder, ) @@ -1444,16 +1445,24 @@ class FederationHandler(BaseHandler): auth_events=auth_events, ) - if not event.internal_metadata.is_outlier() and not backfilled: - yield self.action_generator.handle_push_actions_for_event( - event, context - ) + try: + if not event.internal_metadata.is_outlier() and not backfilled: + yield self.action_generator.handle_push_actions_for_event( + event, context + ) - event_stream_id, max_stream_id = yield self.store.persist_event( - event, - context=context, - backfilled=backfilled, - ) + event_stream_id, max_stream_id = yield self.store.persist_event( + event, + context=context, + backfilled=backfilled, + ) + except: # noqa: E722, as we reraise the exception this is fine. + # Ensure that we actually remove the entries in the push actions + # staging area + logcontext.preserve_fn( + self.store.remove_push_actions_from_staging + )(event.event_id) + raise if not backfilled: # this intentionally does not yield: we don't care about the result @@ -1828,8 +1837,8 @@ class FederationHandler(BaseHandler): current_state = set(e.event_id for e in auth_events.values()) different_auth = event_auth_events - current_state - self._update_context_for_auth_events( - context, auth_events, event_key, + yield self._update_context_for_auth_events( + event, context, auth_events, event_key, ) if different_auth and not event.internal_metadata.is_outlier(): @@ -1910,8 +1919,8 @@ class FederationHandler(BaseHandler): # 4. Look at rejects and their proofs. # TODO. - self._update_context_for_auth_events( - context, auth_events, event_key, + yield self._update_context_for_auth_events( + event, context, auth_events, event_key, ) try: @@ -1920,11 +1929,15 @@ class FederationHandler(BaseHandler): logger.warn("Failed auth resolution for %r because %s", event, e) raise e - def _update_context_for_auth_events(self, context, auth_events, + @defer.inlineCallbacks + def _update_context_for_auth_events(self, event, context, auth_events, event_key): - """Update the state_ids in an event context after auth event resolution + """Update the state_ids in an event context after auth event resolution, + storing the changes as a new state group. Args: + event (Event): The event we're handling the context for + context (synapse.events.snapshot.EventContext): event context to be updated @@ -1947,7 +1960,13 @@ class FederationHandler(BaseHandler): context.prev_state_ids.update({ k: a.event_id for k, a in auth_events.iteritems() }) - context.state_group = self.store.get_next_state_group() + context.state_group = yield self.store.store_state_group( + event.event_id, + event.room_id, + prev_group=context.prev_group, + delta_ids=context.delta_ids, + current_state_ids=context.current_state_ids, + ) @defer.inlineCallbacks def construct_auth_difference(self, local_auth, remote_auth): @@ -2117,8 +2136,7 @@ class FederationHandler(BaseHandler): if (yield self.auth.check_host_in_room(room_id, self.hs.hostname)): builder = self.event_builder_factory.new(event_dict) EventValidator().validate_new(builder) - message_handler = self.hs.get_handlers().message_handler - event, context = yield message_handler._create_new_client_event( + event, context = yield self.event_creation_handler.create_new_client_event( builder=builder ) @@ -2133,7 +2151,7 @@ class FederationHandler(BaseHandler): raise e yield self._check_signature(event, context) - member_handler = self.hs.get_handlers().room_member_handler + member_handler = self.hs.get_room_member_handler() yield member_handler.send_membership_event(None, event, context) else: destinations = set(x.split(":", 1)[-1] for x in (sender_user_id, room_id)) @@ -2156,8 +2174,7 @@ class FederationHandler(BaseHandler): """ builder = self.event_builder_factory.new(event_dict) - message_handler = self.hs.get_handlers().message_handler - event, context = yield message_handler._create_new_client_event( + event, context = yield self.event_creation_handler.create_new_client_event( builder=builder, ) @@ -2178,7 +2195,7 @@ class FederationHandler(BaseHandler): # TODO: Make sure the signatures actually are correct. event.signatures.update(returned_invite.signatures) - member_handler = self.hs.get_handlers().room_member_handler + member_handler = self.hs.get_room_member_handler() yield member_handler.send_membership_event(None, event, context) @defer.inlineCallbacks @@ -2207,8 +2224,9 @@ class FederationHandler(BaseHandler): builder = self.event_builder_factory.new(event_dict) EventValidator().validate_new(builder) - message_handler = self.hs.get_handlers().message_handler - event, context = yield message_handler._create_new_client_event(builder=builder) + event, context = yield self.event_creation_handler.create_new_client_event( + builder=builder, + ) defer.returnValue((event, context)) @defer.inlineCallbacks diff --git a/synapse/handlers/groups_local.py b/synapse/handlers/groups_local.py index 7e5d3f148d..e4d0cc8b02 100644 --- a/synapse/handlers/groups_local.py +++ b/synapse/handlers/groups_local.py @@ -383,11 +383,12 @@ class GroupsLocalHandler(object): defer.returnValue({"groups": result}) else: - result = yield self.transport_client.get_publicised_groups_for_user( - get_domain_from_id(user_id), user_id + bulk_result = yield self.transport_client.bulk_get_publicised_groups( + get_domain_from_id(user_id), [user_id], ) + result = bulk_result.get("users", {}).get(user_id) # TODO: Verify attestations - defer.returnValue(result) + defer.returnValue({"groups": result}) @defer.inlineCallbacks def bulk_get_publicised_groups(self, user_ids, proxy=True): diff --git a/synapse/handlers/message.py b/synapse/handlers/message.py index d7413833ed..5a8ddc253e 100644 --- a/synapse/handlers/message.py +++ b/synapse/handlers/message.py @@ -1,6 +1,6 @@ # -*- coding: utf-8 -*- # Copyright 2014 - 2016 OpenMarket Ltd -# Copyright 2017 New Vector Ltd +# Copyright 2017 - 2018 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. @@ -13,7 +13,8 @@ # 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 twisted.internet import defer +from twisted.internet import defer, reactor +from twisted.python.failure import Failure from synapse.api.constants import EventTypes, Membership from synapse.api.errors import AuthError, Codes, SynapseError @@ -24,10 +25,12 @@ from synapse.types import ( UserID, RoomAlias, RoomStreamToken, ) from synapse.util.async import run_on_reactor, ReadWriteLock, Limiter -from synapse.util.logcontext import preserve_fn +from synapse.util.logcontext import preserve_fn, run_in_background from synapse.util.metrics import measure_func from synapse.util.frozenutils import unfreeze +from synapse.util.stringutils import random_string from synapse.visibility import filter_events_for_client +from synapse.replication.http.send_event import send_event_to_master from ._base import BaseHandler @@ -40,6 +43,36 @@ import simplejson logger = logging.getLogger(__name__) +class PurgeStatus(object): + """Object tracking the status of a purge request + + This class contains information on the progress of a purge request, for + return by get_purge_status. + + Attributes: + status (int): Tracks whether this request has completed. One of + STATUS_{ACTIVE,COMPLETE,FAILED} + """ + + STATUS_ACTIVE = 0 + STATUS_COMPLETE = 1 + STATUS_FAILED = 2 + + STATUS_TEXT = { + STATUS_ACTIVE: "active", + STATUS_COMPLETE: "complete", + STATUS_FAILED: "failed", + } + + def __init__(self): + self.status = PurgeStatus.STATUS_ACTIVE + + def asdict(self): + return { + "status": PurgeStatus.STATUS_TEXT[self.status] + } + + class MessageHandler(BaseHandler): def __init__(self, hs): @@ -47,32 +80,89 @@ class MessageHandler(BaseHandler): self.hs = hs self.state = hs.get_state_handler() self.clock = hs.get_clock() - self.validator = EventValidator() - self.profile_handler = hs.get_profile_handler() self.pagination_lock = ReadWriteLock() + self._purges_in_progress_by_room = set() + # map from purge id to PurgeStatus + self._purges_by_id = {} - self.pusher_pool = hs.get_pusherpool() + def start_purge_history(self, room_id, topological_ordering, + delete_local_events=False): + """Start off a history purge on a room. - # We arbitrarily limit concurrent event creation for a room to 5. - # This is to stop us from diverging history *too* much. - self.limiter = Limiter(max_count=5) + Args: + room_id (str): The room to purge from - self.action_generator = hs.get_action_generator() + topological_ordering (int): minimum topo ordering to preserve + delete_local_events (bool): True to delete local events as well as + remote ones - self.spam_checker = hs.get_spam_checker() + Returns: + str: unique ID for this purge transaction. + """ + if room_id in self._purges_in_progress_by_room: + raise SynapseError( + 400, + "History purge already in progress for %s" % (room_id, ), + ) + + purge_id = random_string(16) + + # we log the purge_id here so that it can be tied back to the + # request id in the log lines. + logger.info("[purge] starting purge_id %s", purge_id) + + self._purges_by_id[purge_id] = PurgeStatus() + run_in_background( + self._purge_history, + purge_id, room_id, topological_ordering, delete_local_events, + ) + return purge_id @defer.inlineCallbacks - def purge_history(self, room_id, event_id): - event = yield self.store.get_event(event_id) + def _purge_history(self, purge_id, room_id, topological_ordering, + delete_local_events): + """Carry out a history purge on a room. + + Args: + purge_id (str): The id for this purge + room_id (str): The room to purge from + topological_ordering (int): minimum topo ordering to preserve + delete_local_events (bool): True to delete local events as well as + remote ones + + Returns: + Deferred + """ + 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, topological_ordering, delete_local_events, + ) + logger.info("[purge] complete") + self._purges_by_id[purge_id].status = PurgeStatus.STATUS_COMPLETE + except Exception: + logger.error("[purge] failed: %s", Failure().getTraceback().rstrip()) + self._purges_by_id[purge_id].status = PurgeStatus.STATUS_FAILED + finally: + self._purges_in_progress_by_room.discard(room_id) - if event.room_id != room_id: - raise SynapseError(400, "Event is for wrong room.") + # remove the purge from the list 24 hours after it completes + def clear_purge(): + del self._purges_by_id[purge_id] + reactor.callLater(24 * 3600, clear_purge) - depth = event.depth + def get_purge_status(self, purge_id): + """Get the current status of an active purge - with (yield self.pagination_lock.write(room_id)): - yield self.store.delete_old_state(room_id, depth) + Args: + purge_id (str): purge_id returned by start_purge_history + + Returns: + PurgeStatus|None + """ + return self._purges_by_id.get(purge_id) @defer.inlineCallbacks def get_messages(self, requester, room_id=None, pagin_config=None, @@ -183,6 +273,165 @@ class MessageHandler(BaseHandler): defer.returnValue(chunk) @defer.inlineCallbacks + def get_room_data(self, user_id=None, room_id=None, + event_type=None, state_key="", is_guest=False): + """ Get data from a room. + + Args: + event : The room path event + Returns: + The path data content. + Raises: + SynapseError if something went wrong. + """ + membership, membership_event_id = yield self._check_in_room_or_world_readable( + room_id, user_id + ) + + if membership == Membership.JOIN: + data = yield self.state_handler.get_current_state( + room_id, event_type, state_key + ) + elif membership == Membership.LEAVE: + key = (event_type, state_key) + room_state = yield self.store.get_state_for_events( + [membership_event_id], [key] + ) + data = room_state[membership_event_id].get(key) + + defer.returnValue(data) + + @defer.inlineCallbacks + def _check_in_room_or_world_readable(self, room_id, user_id): + try: + # check_user_was_in_room will return the most recent membership + # event for the user if: + # * The user is a non-guest user, and was ever in the room + # * The user is a guest user, and has joined the room + # else it will throw. + member_event = yield self.auth.check_user_was_in_room(room_id, user_id) + defer.returnValue((member_event.membership, member_event.event_id)) + return + except AuthError: + visibility = yield self.state_handler.get_current_state( + room_id, EventTypes.RoomHistoryVisibility, "" + ) + if ( + visibility and + visibility.content["history_visibility"] == "world_readable" + ): + defer.returnValue((Membership.JOIN, None)) + return + raise AuthError( + 403, "Guest access not allowed", errcode=Codes.GUEST_ACCESS_FORBIDDEN + ) + + @defer.inlineCallbacks + def get_state_events(self, user_id, room_id, is_guest=False): + """Retrieve all state events for a given room. If the user is + joined to the room then return the current state. If the user has + left the room return the state events from when they left. + + Args: + user_id(str): The user requesting state events. + room_id(str): The room ID to get all state events from. + Returns: + A list of dicts representing state events. [{}, {}, {}] + """ + membership, membership_event_id = yield self._check_in_room_or_world_readable( + room_id, user_id + ) + + if membership == Membership.JOIN: + room_state = yield self.state_handler.get_current_state(room_id) + elif membership == Membership.LEAVE: + room_state = yield self.store.get_state_for_events( + [membership_event_id], None + ) + room_state = room_state[membership_event_id] + + now = self.clock.time_msec() + defer.returnValue( + [serialize_event(c, now) for c in room_state.values()] + ) + + @defer.inlineCallbacks + def get_joined_members(self, requester, room_id): + """Get all the joined members in the room and their profile information. + + If the user has left the room return the state events from when they left. + + Args: + requester(Requester): The user requesting state events. + room_id(str): The room ID to get all state events from. + Returns: + A dict of user_id to profile info + """ + user_id = requester.user.to_string() + if not requester.app_service: + # We check AS auth after fetching the room membership, as it + # requires us to pull out all joined members anyway. + membership, _ = yield self._check_in_room_or_world_readable( + room_id, user_id + ) + if membership != Membership.JOIN: + raise NotImplementedError( + "Getting joined members after leaving is not implemented" + ) + + users_with_profile = yield self.state.get_current_user_in_room(room_id) + + # If this is an AS, double check that they are allowed to see the members. + # This can either be because the AS user is in the room or becuase there + # is a user in the room that the AS is "interested in" + if requester.app_service and user_id not in users_with_profile: + for uid in users_with_profile: + if requester.app_service.is_interested_in_user(uid): + break + else: + # Loop fell through, AS has no interested users in room + raise AuthError(403, "Appservice not in room") + + defer.returnValue({ + user_id: { + "avatar_url": profile.avatar_url, + "display_name": profile.display_name, + } + for user_id, profile in users_with_profile.iteritems() + }) + + +class EventCreationHandler(object): + def __init__(self, hs): + self.hs = hs + self.auth = hs.get_auth() + self.store = hs.get_datastore() + self.state = hs.get_state_handler() + self.clock = hs.get_clock() + self.validator = EventValidator() + self.profile_handler = hs.get_profile_handler() + self.event_builder_factory = hs.get_event_builder_factory() + self.server_name = hs.hostname + self.ratelimiter = hs.get_ratelimiter() + self.notifier = hs.get_notifier() + self.config = hs.config + + self.http_client = hs.get_simple_http_client() + + # This is only used to get at ratelimit function, and maybe_kick_guest_users + self.base_handler = BaseHandler(hs) + + self.pusher_pool = hs.get_pusherpool() + + # We arbitrarily limit concurrent event creation for a room to 5. + # This is to stop us from diverging history *too* much. + self.limiter = Limiter(max_count=5) + + self.action_generator = hs.get_action_generator() + + self.spam_checker = hs.get_spam_checker() + + @defer.inlineCallbacks def create_event(self, requester, event_dict, token_id=None, txn_id=None, prev_event_ids=None): """ @@ -234,7 +483,7 @@ class MessageHandler(BaseHandler): if txn_id is not None: builder.internal_metadata.txn_id = txn_id - event, context = yield self._create_new_client_event( + event, context = yield self.create_new_client_event( builder=builder, requester=requester, prev_event_ids=prev_event_ids, @@ -259,11 +508,6 @@ class MessageHandler(BaseHandler): "Tried to send member event through non-member codepath" ) - # We check here if we are currently being rate limited, so that we - # don't do unnecessary work. We check again just before we actually - # send the event. - yield self.ratelimit(requester, update=False) - user = UserID.from_string(event.sender) assert self.hs.is_mine(user), "User must be our own: %s" % (user,) @@ -280,12 +524,6 @@ class MessageHandler(BaseHandler): ratelimit=ratelimit, ) - if event.type == EventTypes.Message: - presence = self.hs.get_presence_handler() - # We don't want to block sending messages on any presence code. This - # matters as sometimes presence code can take a while. - preserve_fn(presence.bump_presence_active_time)(user) - @defer.inlineCallbacks def deduplicate_state_event(self, event, context): """ @@ -342,137 +580,9 @@ class MessageHandler(BaseHandler): ) defer.returnValue(event) + @measure_func("create_new_client_event") @defer.inlineCallbacks - def get_room_data(self, user_id=None, room_id=None, - event_type=None, state_key="", is_guest=False): - """ Get data from a room. - - Args: - event : The room path event - Returns: - The path data content. - Raises: - SynapseError if something went wrong. - """ - membership, membership_event_id = yield self._check_in_room_or_world_readable( - room_id, user_id - ) - - if membership == Membership.JOIN: - data = yield self.state_handler.get_current_state( - room_id, event_type, state_key - ) - elif membership == Membership.LEAVE: - key = (event_type, state_key) - room_state = yield self.store.get_state_for_events( - [membership_event_id], [key] - ) - data = room_state[membership_event_id].get(key) - - defer.returnValue(data) - - @defer.inlineCallbacks - def _check_in_room_or_world_readable(self, room_id, user_id): - try: - # check_user_was_in_room will return the most recent membership - # event for the user if: - # * The user is a non-guest user, and was ever in the room - # * The user is a guest user, and has joined the room - # else it will throw. - member_event = yield self.auth.check_user_was_in_room(room_id, user_id) - defer.returnValue((member_event.membership, member_event.event_id)) - return - except AuthError: - visibility = yield self.state_handler.get_current_state( - room_id, EventTypes.RoomHistoryVisibility, "" - ) - if ( - visibility and - visibility.content["history_visibility"] == "world_readable" - ): - defer.returnValue((Membership.JOIN, None)) - return - raise AuthError( - 403, "Guest access not allowed", errcode=Codes.GUEST_ACCESS_FORBIDDEN - ) - - @defer.inlineCallbacks - def get_state_events(self, user_id, room_id, is_guest=False): - """Retrieve all state events for a given room. If the user is - joined to the room then return the current state. If the user has - left the room return the state events from when they left. - - Args: - user_id(str): The user requesting state events. - room_id(str): The room ID to get all state events from. - Returns: - A list of dicts representing state events. [{}, {}, {}] - """ - membership, membership_event_id = yield self._check_in_room_or_world_readable( - room_id, user_id - ) - - if membership == Membership.JOIN: - room_state = yield self.state_handler.get_current_state(room_id) - elif membership == Membership.LEAVE: - room_state = yield self.store.get_state_for_events( - [membership_event_id], None - ) - room_state = room_state[membership_event_id] - - now = self.clock.time_msec() - defer.returnValue( - [serialize_event(c, now) for c in room_state.values()] - ) - - @defer.inlineCallbacks - def get_joined_members(self, requester, room_id): - """Get all the joined members in the room and their profile information. - - If the user has left the room return the state events from when they left. - - Args: - requester(Requester): The user requesting state events. - room_id(str): The room ID to get all state events from. - Returns: - A dict of user_id to profile info - """ - user_id = requester.user.to_string() - if not requester.app_service: - # We check AS auth after fetching the room membership, as it - # requires us to pull out all joined members anyway. - membership, _ = yield self._check_in_room_or_world_readable( - room_id, user_id - ) - if membership != Membership.JOIN: - raise NotImplementedError( - "Getting joined members after leaving is not implemented" - ) - - users_with_profile = yield self.state.get_current_user_in_room(room_id) - - # If this is an AS, double check that they are allowed to see the members. - # This can either be because the AS user is in the room or becuase there - # is a user in the room that the AS is "interested in" - if requester.app_service and user_id not in users_with_profile: - for uid in users_with_profile: - if requester.app_service.is_interested_in_user(uid): - break - else: - # Loop fell through, AS has no interested users in room - raise AuthError(403, "Appservice not in room") - - defer.returnValue({ - user_id: { - "avatar_url": profile.avatar_url, - "display_name": profile.display_name, - } - for user_id, profile in users_with_profile.iteritems() - }) - - @measure_func("_create_new_client_event") - @defer.inlineCallbacks - def _create_new_client_event(self, builder, requester=None, prev_event_ids=None): + def create_new_client_event(self, builder, requester=None, prev_event_ids=None): if prev_event_ids: prev_events = yield self.store.add_event_hashes(prev_event_ids) prev_max_depth = yield self.store.get_max_depth_of_events(prev_event_ids) @@ -509,9 +619,7 @@ class MessageHandler(BaseHandler): builder.prev_events = prev_events builder.depth = depth - state_handler = self.state_handler - - context = yield state_handler.compute_event_context(builder) + context = yield self.state.compute_event_context(builder) if requester: context.app_service = requester.app_service @@ -546,12 +654,21 @@ class MessageHandler(BaseHandler): event, context, ratelimit=True, - extra_users=[] + extra_users=[], ): - # We now need to go and hit out to wherever we need to hit out to. + """Processes a new event. This includes checking auth, persisting it, + notifying users, sending to remote servers, etc. - if ratelimit: - yield self.ratelimit(requester) + If called from a worker will hit out to the master process for final + processing. + + Args: + requester (Requester) + event (FrozenEvent) + context (EventContext) + ratelimit (bool) + extra_users (list(UserID)): Any extra users to notify about event + """ try: yield self.auth.check_from_context(event, context) @@ -567,7 +684,58 @@ class MessageHandler(BaseHandler): logger.exception("Failed to encode content: %r", event.content) raise - yield self.maybe_kick_guest_users(event, context) + yield self.action_generator.handle_push_actions_for_event( + event, context + ) + + try: + # If we're a worker we need to hit out to the master. + if self.config.worker_app: + yield send_event_to_master( + self.http_client, + host=self.config.worker_replication_host, + port=self.config.worker_replication_http_port, + requester=requester, + event=event, + context=context, + ratelimit=ratelimit, + extra_users=extra_users, + ) + return + + yield self.persist_and_notify_client_event( + requester, + event, + context, + ratelimit=ratelimit, + extra_users=extra_users, + ) + except: # noqa: E722, as we reraise the exception this is fine. + # Ensure that we actually remove the entries in the push actions + # staging area, if we calculated them. + preserve_fn(self.store.remove_push_actions_from_staging)(event.event_id) + raise + + @defer.inlineCallbacks + def persist_and_notify_client_event( + self, + requester, + event, + context, + ratelimit=True, + extra_users=[], + ): + """Called when we have fully built the event, have already + calculated the push actions for the event, and checked auth. + + This should only be run on master. + """ + assert not self.config.worker_app + + if ratelimit: + yield self.base_handler.ratelimit(requester) + + yield self.base_handler.maybe_kick_guest_users(event, context) if event.type == EventTypes.CanonicalAlias: # Check the alias is acually valid (at this time at least) @@ -660,10 +828,6 @@ class MessageHandler(BaseHandler): "Changing the room create event is forbidden", ) - yield self.action_generator.handle_push_actions_for_event( - event, context - ) - (event_stream_id, max_stream_id) = yield self.store.persist_event( event, context=context ) @@ -683,3 +847,9 @@ class MessageHandler(BaseHandler): ) preserve_fn(_notify)() + + if event.type == EventTypes.Message: + presence = self.hs.get_presence_handler() + # We don't want to block sending messages on any presence code. This + # matters as sometimes presence code can take a while. + preserve_fn(presence.bump_presence_active_time)(requester.user) diff --git a/synapse/handlers/presence.py b/synapse/handlers/presence.py index cb158ba962..a5e501897c 100644 --- a/synapse/handlers/presence.py +++ b/synapse/handlers/presence.py @@ -93,29 +93,30 @@ class PresenceHandler(object): self.store = hs.get_datastore() self.wheel_timer = WheelTimer() self.notifier = hs.get_notifier() - self.replication = hs.get_replication_layer() self.federation = hs.get_federation_sender() self.state = hs.get_state_handler() - self.replication.register_edu_handler( + federation_registry = hs.get_federation_registry() + + federation_registry.register_edu_handler( "m.presence", self.incoming_presence ) - self.replication.register_edu_handler( + federation_registry.register_edu_handler( "m.presence_invite", lambda origin, content: self.invite_presence( observed_user=UserID.from_string(content["observed_user"]), observer_user=UserID.from_string(content["observer_user"]), ) ) - self.replication.register_edu_handler( + federation_registry.register_edu_handler( "m.presence_accept", lambda origin, content: self.accept_presence( observed_user=UserID.from_string(content["observed_user"]), observer_user=UserID.from_string(content["observer_user"]), ) ) - self.replication.register_edu_handler( + federation_registry.register_edu_handler( "m.presence_deny", lambda origin, content: self.deny_presence( observed_user=UserID.from_string(content["observed_user"]), diff --git a/synapse/handlers/profile.py b/synapse/handlers/profile.py index 9800e24453..3465a787ab 100644 --- a/synapse/handlers/profile.py +++ b/synapse/handlers/profile.py @@ -31,14 +31,17 @@ class ProfileHandler(BaseHandler): def __init__(self, hs): super(ProfileHandler, self).__init__(hs) - self.federation = hs.get_replication_layer() - self.federation.register_query_handler( + self.federation = hs.get_federation_client() + hs.get_federation_registry().register_query_handler( "profile", self.on_profile_query ) self.user_directory_handler = hs.get_user_directory_handler() - self.clock.looping_call(self._update_remote_profile_cache, self.PROFILE_UPDATE_MS) + if hs.config.worker_app is None: + self.clock.looping_call( + self._update_remote_profile_cache, self.PROFILE_UPDATE_MS, + ) @defer.inlineCallbacks def get_profile(self, user_id): @@ -233,7 +236,7 @@ class ProfileHandler(BaseHandler): ) for room_id in room_ids: - handler = self.hs.get_handlers().room_member_handler + handler = self.hs.get_room_member_handler() try: # Assume the target_user isn't a guest, # because we don't let guests set profile or avatar data. diff --git a/synapse/handlers/read_marker.py b/synapse/handlers/read_marker.py index b5b0303d54..5142ae153d 100644 --- a/synapse/handlers/read_marker.py +++ b/synapse/handlers/read_marker.py @@ -41,9 +41,9 @@ class ReadMarkerHandler(BaseHandler): """ with (yield self.read_marker_linearizer.queue((room_id, user_id))): - account_data = yield self.store.get_account_data_for_room(user_id, room_id) - - existing_read_marker = account_data.get("m.fully_read", None) + existing_read_marker = yield self.store.get_account_data_for_room_and_type( + user_id, room_id, "m.fully_read", + ) should_update = True diff --git a/synapse/handlers/receipts.py b/synapse/handlers/receipts.py index 0525765272..3f215c2b4e 100644 --- a/synapse/handlers/receipts.py +++ b/synapse/handlers/receipts.py @@ -35,7 +35,7 @@ class ReceiptsHandler(BaseHandler): self.store = hs.get_datastore() self.hs = hs self.federation = hs.get_federation_sender() - hs.get_replication_layer().register_edu_handler( + hs.get_federation_registry().register_edu_handler( "m.receipt", self._received_remote_receipt ) self.clock = self.hs.get_clock() diff --git a/synapse/handlers/register.py b/synapse/handlers/register.py index 4bc6ef51fe..ed5939880a 100644 --- a/synapse/handlers/register.py +++ b/synapse/handlers/register.py @@ -25,6 +25,7 @@ from synapse.http.client import CaptchaServerHttpClient from synapse import types from synapse.types import UserID from synapse.util.async import run_on_reactor +from synapse.util.threepids import check_3pid_allowed from ._base import BaseHandler logger = logging.getLogger(__name__) @@ -131,7 +132,7 @@ class RegistrationHandler(BaseHandler): yield run_on_reactor() password_hash = None if password: - password_hash = self.auth_handler().hash(password) + password_hash = yield self.auth_handler().hash(password) if localpart: yield self.check_username(localpart, guest_access_token=guest_access_token) @@ -293,7 +294,7 @@ class RegistrationHandler(BaseHandler): """ for c in threepidCreds: - logger.info("validating theeepidcred sid %s on id server %s", + logger.info("validating threepidcred sid %s on id server %s", c['sid'], c['idServer']) try: identity_handler = self.hs.get_handlers().identity_handler @@ -307,6 +308,11 @@ class RegistrationHandler(BaseHandler): logger.info("got threepid with medium '%s' and address '%s'", threepid['medium'], threepid['address']) + if not check_3pid_allowed(self.hs, threepid['medium'], threepid['address']): + raise RegistrationError( + 403, "Third party identifier is not allowed" + ) + @defer.inlineCallbacks def bind_emails(self, user_id, threepidCreds): """Links emails with a user ID and informs an identity server. @@ -440,16 +446,34 @@ class RegistrationHandler(BaseHandler): return self.hs.get_auth_handler() @defer.inlineCallbacks - def guest_access_token_for(self, medium, address, inviter_user_id): + def get_or_register_3pid_guest(self, medium, address, inviter_user_id): + """Get a guest access token for a 3PID, creating a guest account if + one doesn't already exist. + + Args: + medium (str) + address (str) + inviter_user_id (str): The user ID who is trying to invite the + 3PID + + Returns: + Deferred[(str, str)]: A 2-tuple of `(user_id, access_token)` of the + 3PID guest account. + """ access_token = yield self.store.get_3pid_guest_access_token(medium, address) if access_token: - defer.returnValue(access_token) + user_info = yield self.auth.get_user_by_access_token( + access_token + ) - _, access_token = yield self.register( + defer.returnValue((user_info["user"].to_string(), access_token)) + + user_id, access_token = yield self.register( generate_token=True, make_guest=True ) access_token = yield self.store.save_or_get_3pid_guest_access_token( medium, address, access_token, inviter_user_id ) - defer.returnValue(access_token) + + defer.returnValue((user_id, access_token)) diff --git a/synapse/handlers/room.py b/synapse/handlers/room.py index d1cc87a016..8df8fcbbad 100644 --- a/synapse/handlers/room.py +++ b/synapse/handlers/room.py @@ -1,5 +1,6 @@ # -*- coding: utf-8 -*- # Copyright 2014 - 2016 OpenMarket Ltd +# Copyright 2018 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. @@ -64,6 +65,7 @@ class RoomCreationHandler(BaseHandler): super(RoomCreationHandler, self).__init__(hs) self.spam_checker = hs.get_spam_checker() + self.event_creation_handler = hs.get_event_creation_handler() @defer.inlineCallbacks def create_room(self, requester, config, ratelimit=True): @@ -163,13 +165,11 @@ class RoomCreationHandler(BaseHandler): creation_content = config.get("creation_content", {}) - msg_handler = self.hs.get_handlers().message_handler - room_member_handler = self.hs.get_handlers().room_member_handler + room_member_handler = self.hs.get_room_member_handler() yield self._send_events_for_new_room( requester, room_id, - msg_handler, room_member_handler, preset_config=preset_config, invite_list=invite_list, @@ -181,7 +181,7 @@ class RoomCreationHandler(BaseHandler): if "name" in config: name = config["name"] - yield msg_handler.create_and_send_nonmember_event( + yield self.event_creation_handler.create_and_send_nonmember_event( requester, { "type": EventTypes.Name, @@ -194,7 +194,7 @@ class RoomCreationHandler(BaseHandler): if "topic" in config: topic = config["topic"] - yield msg_handler.create_and_send_nonmember_event( + yield self.event_creation_handler.create_and_send_nonmember_event( requester, { "type": EventTypes.Topic, @@ -224,7 +224,7 @@ class RoomCreationHandler(BaseHandler): id_server = invite_3pid["id_server"] address = invite_3pid["address"] medium = invite_3pid["medium"] - yield self.hs.get_handlers().room_member_handler.do_3pid_invite( + yield self.hs.get_room_member_handler().do_3pid_invite( room_id, requester.user, medium, @@ -249,7 +249,6 @@ class RoomCreationHandler(BaseHandler): self, creator, # A Requester object. room_id, - msg_handler, room_member_handler, preset_config, invite_list, @@ -272,7 +271,7 @@ class RoomCreationHandler(BaseHandler): @defer.inlineCallbacks def send(etype, content, **kwargs): event = create(etype, content, **kwargs) - yield msg_handler.create_and_send_nonmember_event( + yield self.event_creation_handler.create_and_send_nonmember_event( creator, event, ratelimit=False @@ -476,12 +475,9 @@ class RoomEventSource(object): user.to_string() ) if app_service: - events, end_key = yield self.store.get_appservice_room_stream( - service=app_service, - from_key=from_key, - to_key=to_key, - limit=limit, - ) + # We no longer support AS users using /sync directly. + # See https://github.com/matrix-org/matrix-doc/issues/1144 + raise NotImplementedError() else: room_events = yield self.store.get_membership_changes_for_user( user.to_string(), from_key, to_key diff --git a/synapse/handlers/room_list.py b/synapse/handlers/room_list.py index bb40075387..5d81f59b44 100644 --- a/synapse/handlers/room_list.py +++ b/synapse/handlers/room_list.py @@ -203,7 +203,8 @@ class RoomListHandler(BaseHandler): if limit: step = limit + 1 else: - step = len(rooms_to_scan) + # step cannot be zero + step = len(rooms_to_scan) if len(rooms_to_scan) != 0 else 1 chunk = [] for i in xrange(0, len(rooms_to_scan), step): @@ -408,7 +409,7 @@ class RoomListHandler(BaseHandler): def _get_remote_list_cached(self, server_name, limit=None, since_token=None, search_filter=None, include_all_networks=False, third_party_instance_id=None,): - repl_layer = self.hs.get_replication_layer() + repl_layer = self.hs.get_federation_client() if search_filter: # We can't cache when asking for search return repl_layer.get_public_rooms( diff --git a/synapse/handlers/room_member.py b/synapse/handlers/room_member.py index 7e6467cd1d..9977be8831 100644 --- a/synapse/handlers/room_member.py +++ b/synapse/handlers/room_member.py @@ -1,5 +1,6 @@ # -*- coding: utf-8 -*- # Copyright 2016 OpenMarket Ltd +# Copyright 2018 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. @@ -13,7 +14,7 @@ # See the License for the specific language governing permissions and # limitations under the License. - +import abc import logging from signedjson.key import decode_verify_key_bytes @@ -29,32 +30,121 @@ from synapse.api.errors import AuthError, SynapseError, Codes from synapse.types import UserID, RoomID from synapse.util.async import Linearizer from synapse.util.distributor import user_left_room, user_joined_room -from ._base import BaseHandler + logger = logging.getLogger(__name__) id_server_scheme = "https://" -class RoomMemberHandler(BaseHandler): +class RoomMemberHandler(object): # TODO(paul): This handler currently contains a messy conflation of # low-level API that works on UserID objects and so on, and REST-level # API that takes ID strings and returns pagination chunks. These concerns # ought to be separated out a lot better. - def __init__(self, hs): - super(RoomMemberHandler, self).__init__(hs) + __metaclass__ = abc.ABCMeta + def __init__(self, hs): + self.hs = hs + self.store = hs.get_datastore() + self.auth = hs.get_auth() + self.state_handler = hs.get_state_handler() + self.config = hs.config + self.simple_http_client = hs.get_simple_http_client() + + self.federation_handler = hs.get_handlers().federation_handler + self.directory_handler = hs.get_handlers().directory_handler + self.registration_handler = hs.get_handlers().registration_handler self.profile_handler = hs.get_profile_handler() + self.event_creation_hander = hs.get_event_creation_handler() self.member_linearizer = Linearizer(name="member") self.clock = hs.get_clock() self.spam_checker = hs.get_spam_checker() - self.distributor = hs.get_distributor() - self.distributor.declare("user_joined_room") - self.distributor.declare("user_left_room") + @abc.abstractmethod + def _remote_join(self, requester, remote_room_hosts, room_id, user, content): + """Try and join a room that this server is not in + + Args: + requester (Requester) + remote_room_hosts (list[str]): List of servers that can be used + to join via. + room_id (str): Room that we are trying to join + user (UserID): User who is trying to join + content (dict): A dict that should be used as the content of the + join event. + + Returns: + Deferred + """ + raise NotImplementedError() + + @abc.abstractmethod + def _remote_reject_invite(self, remote_room_hosts, room_id, target): + """Attempt to reject an invite for a room this server is not in. If we + fail to do so we locally mark the invite as rejected. + + Args: + requester (Requester) + remote_room_hosts (list[str]): List of servers to use to try and + reject invite + room_id (str) + target (UserID): The user rejecting the invite + + Returns: + Deferred[dict]: A dictionary to be returned to the client, may + include event_id etc, or nothing if we locally rejected + """ + raise NotImplementedError() + + @abc.abstractmethod + def get_or_register_3pid_guest(self, requester, medium, address, inviter_user_id): + """Get a guest access token for a 3PID, creating a guest account if + one doesn't already exist. + + Args: + requester (Requester) + medium (str) + address (str) + inviter_user_id (str): The user ID who is trying to invite the + 3PID + + Returns: + Deferred[(str, str)]: A 2-tuple of `(user_id, access_token)` of the + 3PID guest account. + """ + raise NotImplementedError() + + @abc.abstractmethod + def _user_joined_room(self, target, room_id): + """Notifies distributor on master process that the user has joined the + room. + + Args: + target (UserID) + room_id (str) + + Returns: + Deferred|None + """ + raise NotImplementedError() + + @abc.abstractmethod + def _user_left_room(self, target, room_id): + """Notifies distributor on master process that the user has left the + room. + + Args: + target (UserID) + room_id (str) + + Returns: + Deferred|None + """ + raise NotImplementedError() @defer.inlineCallbacks def _local_membership_update( @@ -66,13 +156,12 @@ class RoomMemberHandler(BaseHandler): ): if content is None: content = {} - msg_handler = self.hs.get_handlers().message_handler content["membership"] = membership if requester.is_guest: content["kind"] = "guest" - event, context = yield msg_handler.create_event( + event, context = yield self.event_creation_hander.create_event( requester, { "type": EventTypes.Member, @@ -90,12 +179,14 @@ class RoomMemberHandler(BaseHandler): ) # Check if this event matches the previous membership event for the user. - duplicate = yield msg_handler.deduplicate_state_event(event, context) + duplicate = yield self.event_creation_hander.deduplicate_state_event( + event, context, + ) if duplicate is not None: # Discard the new event since this membership change is a no-op. defer.returnValue(duplicate) - yield msg_handler.handle_new_client_event( + yield self.event_creation_hander.handle_new_client_event( requester, event, context, @@ -117,33 +208,16 @@ class RoomMemberHandler(BaseHandler): prev_member_event = yield self.store.get_event(prev_member_event_id) newly_joined = prev_member_event.membership != Membership.JOIN if newly_joined: - yield user_joined_room(self.distributor, target, room_id) + yield self._user_joined_room(target, room_id) elif event.membership == Membership.LEAVE: if prev_member_event_id: prev_member_event = yield self.store.get_event(prev_member_event_id) if prev_member_event.membership == Membership.JOIN: - user_left_room(self.distributor, target, room_id) + yield self._user_left_room(target, room_id) defer.returnValue(event) @defer.inlineCallbacks - def remote_join(self, remote_room_hosts, room_id, user, content): - if len(remote_room_hosts) == 0: - raise SynapseError(404, "No known servers") - - # We don't do an auth check if we are doing an invite - # join dance for now, since we're kinda implicitly checking - # that we are allowed to join when we decide whether or not we - # need to do the invite/join dance. - yield self.hs.get_handlers().federation_handler.do_invite_join( - remote_room_hosts, - room_id, - user.to_string(), - content, - ) - yield user_joined_room(self.distributor, user, room_id) - - @defer.inlineCallbacks def update_membership( self, requester, @@ -201,8 +275,7 @@ class RoomMemberHandler(BaseHandler): # if this is a join with a 3pid signature, we may need to turn a 3pid # invite into a normal invite before we can handle the join. if third_party_signed is not None: - replication = self.hs.get_replication_layer() - yield replication.exchange_third_party_invite( + yield self.federation_handler.exchange_third_party_invite( third_party_signed["sender"], target.to_string(), room_id, @@ -223,7 +296,7 @@ class RoomMemberHandler(BaseHandler): requester.user, ) if not is_requester_admin: - if self.hs.config.block_non_admin_invites: + if self.config.block_non_admin_invites: logger.info( "Blocking invite: user is not admin and non-admin " "invites disabled" @@ -282,7 +355,7 @@ class RoomMemberHandler(BaseHandler): raise AuthError(403, "Guest access not allowed") if not is_host_in_room: - inviter = yield self.get_inviter(target.to_string(), room_id) + inviter = yield self._get_inviter(target.to_string(), room_id) if inviter and not self.hs.is_mine(inviter): remote_room_hosts.append(inviter.domain) @@ -296,15 +369,15 @@ class RoomMemberHandler(BaseHandler): if requester.is_guest: content["kind"] = "guest" - ret = yield self.remote_join( - remote_room_hosts, room_id, target, content + ret = yield self._remote_join( + requester, remote_room_hosts, room_id, target, content ) defer.returnValue(ret) elif effective_membership_state == Membership.LEAVE: if not is_host_in_room: # perhaps we've been invited - inviter = yield self.get_inviter(target.to_string(), room_id) + inviter = yield self._get_inviter(target.to_string(), room_id) if not inviter: raise SynapseError(404, "Not a known room") @@ -318,28 +391,10 @@ class RoomMemberHandler(BaseHandler): else: # send the rejection to the inviter's HS. remote_room_hosts = remote_room_hosts + [inviter.domain] - fed_handler = self.hs.get_handlers().federation_handler - try: - ret = yield fed_handler.do_remotely_reject_invite( - remote_room_hosts, - room_id, - target.to_string(), - ) - defer.returnValue(ret) - except Exception as e: - # if we were unable to reject the exception, just mark - # it as rejected on our end and plough ahead. - # - # The 'except' clause is very broad, but we need to - # capture everything from DNS failures upwards - # - logger.warn("Failed to reject invite: %s", e) - - yield self.store.locally_reject_invite( - target.to_string(), room_id - ) - - defer.returnValue({}) + res = yield self._remote_reject_invite( + requester, remote_room_hosts, room_id, target, + ) + defer.returnValue(res) res = yield self._local_membership_update( requester=requester, @@ -394,8 +449,9 @@ class RoomMemberHandler(BaseHandler): else: requester = synapse.types.create_requester(target_user) - message_handler = self.hs.get_handlers().message_handler - prev_event = yield message_handler.deduplicate_state_event(event, context) + prev_event = yield self.event_creation_hander.deduplicate_state_event( + event, context, + ) if prev_event is not None: return @@ -412,7 +468,7 @@ class RoomMemberHandler(BaseHandler): if is_blocked: raise SynapseError(403, "This room has been blocked on this server") - yield message_handler.handle_new_client_event( + yield self.event_creation_hander.handle_new_client_event( requester, event, context, @@ -434,12 +490,12 @@ class RoomMemberHandler(BaseHandler): prev_member_event = yield self.store.get_event(prev_member_event_id) newly_joined = prev_member_event.membership != Membership.JOIN if newly_joined: - yield user_joined_room(self.distributor, target_user, room_id) + yield self._user_joined_room(target_user, room_id) elif event.membership == Membership.LEAVE: if prev_member_event_id: prev_member_event = yield self.store.get_event(prev_member_event_id) if prev_member_event.membership == Membership.JOIN: - user_left_room(self.distributor, target_user, room_id) + yield self._user_left_room(target_user, room_id) @defer.inlineCallbacks def _can_guest_join(self, current_state_ids): @@ -473,7 +529,7 @@ class RoomMemberHandler(BaseHandler): Raises: SynapseError if room alias could not be found. """ - directory_handler = self.hs.get_handlers().directory_handler + directory_handler = self.directory_handler mapping = yield directory_handler.get_association(room_alias) if not mapping: @@ -485,7 +541,7 @@ class RoomMemberHandler(BaseHandler): defer.returnValue((RoomID.from_string(room_id), servers)) @defer.inlineCallbacks - def get_inviter(self, user_id, room_id): + def _get_inviter(self, user_id, room_id): invite = yield self.store.get_invite_for_user_in_room( user_id=user_id, room_id=room_id, @@ -504,7 +560,7 @@ class RoomMemberHandler(BaseHandler): requester, txn_id ): - if self.hs.config.block_non_admin_invites: + if self.config.block_non_admin_invites: is_requester_admin = yield self.auth.is_server_admin( requester.user, ) @@ -551,7 +607,7 @@ class RoomMemberHandler(BaseHandler): str: the matrix ID of the 3pid, or None if it is not recognized. """ try: - data = yield self.hs.get_simple_http_client().get_json( + data = yield self.simple_http_client.get_json( "%s%s/_matrix/identity/api/v1/lookup" % (id_server_scheme, id_server,), { "medium": medium, @@ -562,7 +618,7 @@ class RoomMemberHandler(BaseHandler): if "mxid" in data: if "signatures" not in data: raise AuthError(401, "No signatures on 3pid binding") - self.verify_any_signature(data, id_server) + yield self._verify_any_signature(data, id_server) defer.returnValue(data["mxid"]) except IOError as e: @@ -570,11 +626,11 @@ class RoomMemberHandler(BaseHandler): defer.returnValue(None) @defer.inlineCallbacks - def verify_any_signature(self, data, server_hostname): + def _verify_any_signature(self, data, server_hostname): if server_hostname not in data["signatures"]: raise AuthError(401, "No signature from server %s" % (server_hostname,)) for key_name, signature in data["signatures"][server_hostname].items(): - key_data = yield self.hs.get_simple_http_client().get_json( + key_data = yield self.simple_http_client.get_json( "%s%s/_matrix/identity/api/v1/pubkey/%s" % (id_server_scheme, server_hostname, key_name,), ) @@ -599,7 +655,7 @@ class RoomMemberHandler(BaseHandler): user, txn_id ): - room_state = yield self.hs.get_state_handler().get_current_state(room_id) + room_state = yield self.state_handler.get_current_state(room_id) inviter_display_name = "" inviter_avatar_url = "" @@ -630,6 +686,7 @@ class RoomMemberHandler(BaseHandler): token, public_keys, fallback_public_key, display_name = ( yield self._ask_id_server_for_third_party_invite( + requester=requester, id_server=id_server, medium=medium, address=address, @@ -644,8 +701,7 @@ class RoomMemberHandler(BaseHandler): ) ) - msg_handler = self.hs.get_handlers().message_handler - yield msg_handler.create_and_send_nonmember_event( + yield self.event_creation_hander.create_and_send_nonmember_event( requester, { "type": EventTypes.ThirdPartyInvite, @@ -667,6 +723,7 @@ class RoomMemberHandler(BaseHandler): @defer.inlineCallbacks def _ask_id_server_for_third_party_invite( self, + requester, id_server, medium, address, @@ -683,6 +740,7 @@ class RoomMemberHandler(BaseHandler): Asks an identity server for a third party invite. Args: + requester (Requester) id_server (str): hostname + optional port for the identity server. medium (str): The literal string "email". address (str): The third party address being invited. @@ -724,24 +782,20 @@ class RoomMemberHandler(BaseHandler): "sender_avatar_url": inviter_avatar_url, } - if self.hs.config.invite_3pid_guest: - registration_handler = self.hs.get_handlers().registration_handler - guest_access_token = yield registration_handler.guest_access_token_for( + if self.config.invite_3pid_guest: + guest_access_token, guest_user_id = yield self.get_or_register_3pid_guest( + requester=requester, medium=medium, address=address, inviter_user_id=inviter_user_id, ) - guest_user_info = yield self.hs.get_auth().get_user_by_access_token( - guest_access_token - ) - invite_config.update({ "guest_access_token": guest_access_token, - "guest_user_id": guest_user_info["user"].to_string(), + "guest_user_id": guest_user_id, }) - data = yield self.hs.get_simple_http_client().post_urlencoded_get_json( + data = yield self.simple_http_client.post_urlencoded_get_json( is_url, invite_config ) @@ -764,27 +818,6 @@ class RoomMemberHandler(BaseHandler): defer.returnValue((token, public_keys, fallback_public_key, display_name)) @defer.inlineCallbacks - def forget(self, user, room_id): - user_id = user.to_string() - - member = yield self.state_handler.get_current_state( - room_id=room_id, - event_type=EventTypes.Member, - state_key=user_id - ) - membership = member.membership if member else None - - if membership is not None and membership not in [ - Membership.LEAVE, Membership.BAN - ]: - raise SynapseError(400, "User %s in room %s" % ( - user_id, room_id - )) - - if membership: - yield self.store.forget(user_id, room_id) - - @defer.inlineCallbacks def _is_host_in_room(self, current_state_ids): # Have we just created the room, and is this about to be the very # first member event? @@ -805,3 +838,94 @@ class RoomMemberHandler(BaseHandler): defer.returnValue(True) defer.returnValue(False) + + +class RoomMemberMasterHandler(RoomMemberHandler): + def __init__(self, hs): + super(RoomMemberMasterHandler, self).__init__(hs) + + self.distributor = hs.get_distributor() + self.distributor.declare("user_joined_room") + self.distributor.declare("user_left_room") + + @defer.inlineCallbacks + def _remote_join(self, requester, remote_room_hosts, room_id, user, content): + """Implements RoomMemberHandler._remote_join + """ + if len(remote_room_hosts) == 0: + raise SynapseError(404, "No known servers") + + # We don't do an auth check if we are doing an invite + # join dance for now, since we're kinda implicitly checking + # that we are allowed to join when we decide whether or not we + # need to do the invite/join dance. + yield self.federation_handler.do_invite_join( + remote_room_hosts, + room_id, + user.to_string(), + content, + ) + yield self._user_joined_room(user, room_id) + + @defer.inlineCallbacks + def _remote_reject_invite(self, requester, remote_room_hosts, room_id, target): + """Implements RoomMemberHandler._remote_reject_invite + """ + fed_handler = self.federation_handler + try: + ret = yield fed_handler.do_remotely_reject_invite( + remote_room_hosts, + room_id, + target.to_string(), + ) + defer.returnValue(ret) + except Exception as e: + # if we were unable to reject the exception, just mark + # it as rejected on our end and plough ahead. + # + # The 'except' clause is very broad, but we need to + # capture everything from DNS failures upwards + # + logger.warn("Failed to reject invite: %s", e) + + yield self.store.locally_reject_invite( + target.to_string(), room_id + ) + defer.returnValue({}) + + def get_or_register_3pid_guest(self, requester, medium, address, inviter_user_id): + """Implements RoomMemberHandler.get_or_register_3pid_guest + """ + rg = self.registration_handler + return rg.get_or_register_3pid_guest(medium, address, inviter_user_id) + + def _user_joined_room(self, target, room_id): + """Implements RoomMemberHandler._user_joined_room + """ + return user_joined_room(self.distributor, target, room_id) + + def _user_left_room(self, target, room_id): + """Implements RoomMemberHandler._user_left_room + """ + return user_left_room(self.distributor, target, room_id) + + @defer.inlineCallbacks + def forget(self, user, room_id): + user_id = user.to_string() + + member = yield self.state_handler.get_current_state( + room_id=room_id, + event_type=EventTypes.Member, + state_key=user_id + ) + membership = member.membership if member else None + + if membership is not None and membership not in [ + Membership.LEAVE, Membership.BAN + ]: + raise SynapseError(400, "User %s in room %s" % ( + user_id, room_id + )) + + if membership: + yield self.store.forget(user_id, room_id) diff --git a/synapse/handlers/room_member_worker.py b/synapse/handlers/room_member_worker.py new file mode 100644 index 0000000000..493aec1e48 --- /dev/null +++ b/synapse/handlers/room_member_worker.py @@ -0,0 +1,102 @@ +# -*- coding: utf-8 -*- +# Copyright 2018 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. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +import logging + +from twisted.internet import defer + +from synapse.api.errors import SynapseError +from synapse.handlers.room_member import RoomMemberHandler +from synapse.replication.http.membership import ( + remote_join, remote_reject_invite, get_or_register_3pid_guest, + notify_user_membership_change, +) + + +logger = logging.getLogger(__name__) + + +class RoomMemberWorkerHandler(RoomMemberHandler): + @defer.inlineCallbacks + def _remote_join(self, requester, remote_room_hosts, room_id, user, content): + """Implements RoomMemberHandler._remote_join + """ + if len(remote_room_hosts) == 0: + raise SynapseError(404, "No known servers") + + ret = yield remote_join( + self.simple_http_client, + host=self.config.worker_replication_host, + port=self.config.worker_replication_http_port, + requester=requester, + remote_room_hosts=remote_room_hosts, + room_id=room_id, + user_id=user.to_string(), + content=content, + ) + + yield self._user_joined_room(user, room_id) + + defer.returnValue(ret) + + def _remote_reject_invite(self, requester, remote_room_hosts, room_id, target): + """Implements RoomMemberHandler._remote_reject_invite + """ + return remote_reject_invite( + self.simple_http_client, + host=self.config.worker_replication_host, + port=self.config.worker_replication_http_port, + requester=requester, + remote_room_hosts=remote_room_hosts, + room_id=room_id, + user_id=target.to_string(), + ) + + def _user_joined_room(self, target, room_id): + """Implements RoomMemberHandler._user_joined_room + """ + return notify_user_membership_change( + self.simple_http_client, + host=self.config.worker_replication_host, + port=self.config.worker_replication_http_port, + user_id=target.to_string(), + room_id=room_id, + change="joined", + ) + + def _user_left_room(self, target, room_id): + """Implements RoomMemberHandler._user_left_room + """ + return notify_user_membership_change( + self.simple_http_client, + host=self.config.worker_replication_host, + port=self.config.worker_replication_http_port, + user_id=target.to_string(), + room_id=room_id, + change="left", + ) + + def get_or_register_3pid_guest(self, requester, medium, address, inviter_user_id): + """Implements RoomMemberHandler.get_or_register_3pid_guest + """ + return get_or_register_3pid_guest( + self.simple_http_client, + host=self.config.worker_replication_host, + port=self.config.worker_replication_http_port, + requester=requester, + medium=medium, + address=address, + inviter_user_id=inviter_user_id, + ) diff --git a/synapse/handlers/set_password.py b/synapse/handlers/set_password.py index 44414e1dc1..e057ae54c9 100644 --- a/synapse/handlers/set_password.py +++ b/synapse/handlers/set_password.py @@ -31,7 +31,7 @@ class SetPasswordHandler(BaseHandler): @defer.inlineCallbacks def set_password(self, user_id, newpassword, requester=None): - password_hash = self._auth_handler.hash(newpassword) + password_hash = yield self._auth_handler.hash(newpassword) except_device_id = requester.device_id if requester else None except_access_token_id = requester.access_token_id if requester else None diff --git a/synapse/handlers/sync.py b/synapse/handlers/sync.py index b12988f3c9..0f713ce038 100644 --- a/synapse/handlers/sync.py +++ b/synapse/handlers/sync.py @@ -235,10 +235,10 @@ class SyncHandler(object): defer.returnValue(rules) @defer.inlineCallbacks - def ephemeral_by_room(self, sync_config, now_token, since_token=None): + def ephemeral_by_room(self, sync_result_builder, now_token, since_token=None): """Get the ephemeral events for each room the user is in Args: - sync_config (SyncConfig): The flags, filters and user for the sync. + sync_result_builder(SyncResultBuilder) now_token (StreamToken): Where the server is currently up to. since_token (StreamToken): Where the server was when the client last synced. @@ -248,10 +248,12 @@ class SyncHandler(object): typing events for that room. """ + sync_config = sync_result_builder.sync_config + with Measure(self.clock, "ephemeral_by_room"): typing_key = since_token.typing_key if since_token else "0" - room_ids = yield self.store.get_rooms_for_user(sync_config.user.to_string()) + room_ids = sync_result_builder.joined_room_ids typing_source = self.event_sources.sources["typing"] typing, typing_key = yield typing_source.get_new_events( @@ -565,10 +567,22 @@ class SyncHandler(object): # Always use the `now_token` in `SyncResultBuilder` now_token = yield self.event_sources.get_current_token() + user_id = sync_config.user.to_string() + app_service = self.store.get_app_service_by_user_id(user_id) + if app_service: + # We no longer support AS users using /sync directly. + # See https://github.com/matrix-org/matrix-doc/issues/1144 + raise NotImplementedError() + else: + joined_room_ids = yield self.get_rooms_for_user_at( + user_id, now_token.room_stream_id, + ) + sync_result_builder = SyncResultBuilder( sync_config, full_state, since_token=since_token, now_token=now_token, + joined_room_ids=joined_room_ids, ) account_data_by_room = yield self._generate_sync_entry_for_account_data( @@ -603,7 +617,6 @@ class SyncHandler(object): device_id = sync_config.device_id one_time_key_counts = {} if device_id: - user_id = sync_config.user.to_string() one_time_key_counts = yield self.store.count_e2e_one_time_keys( user_id, device_id ) @@ -891,7 +904,7 @@ class SyncHandler(object): ephemeral_by_room = {} else: now_token, ephemeral_by_room = yield self.ephemeral_by_room( - sync_result_builder.sync_config, + sync_result_builder, now_token=sync_result_builder.now_token, since_token=sync_result_builder.since_token, ) @@ -996,15 +1009,8 @@ class SyncHandler(object): if rooms_changed: defer.returnValue(True) - app_service = self.store.get_app_service_by_user_id(user_id) - if app_service: - rooms = yield self.store.get_app_service_rooms(app_service) - joined_room_ids = set(r.room_id for r in rooms) - else: - joined_room_ids = yield self.store.get_rooms_for_user(user_id) - stream_id = RoomStreamToken.parse_stream_token(since_token.room_key).stream - for room_id in joined_room_ids: + for room_id in sync_result_builder.joined_room_ids: if self.store.has_room_changed_since(room_id, stream_id): defer.returnValue(True) defer.returnValue(False) @@ -1028,13 +1034,6 @@ class SyncHandler(object): assert since_token - app_service = self.store.get_app_service_by_user_id(user_id) - if app_service: - rooms = yield self.store.get_app_service_rooms(app_service) - joined_room_ids = set(r.room_id for r in rooms) - else: - joined_room_ids = yield self.store.get_rooms_for_user(user_id) - # Get a list of membership change events that have happened. rooms_changed = yield self.store.get_membership_changes_for_user( user_id, since_token.room_key, now_token.room_key @@ -1057,7 +1056,7 @@ class SyncHandler(object): # we do send down the room, and with full state, where necessary old_state_ids = None - if room_id in joined_room_ids and non_joins: + if room_id in sync_result_builder.joined_room_ids and non_joins: # Always include if the user (re)joined the room, especially # important so that device list changes are calculated correctly. # If there are non join member events, but we are still in the room, @@ -1067,7 +1066,7 @@ class SyncHandler(object): # User is in the room so we don't need to do the invite/leave checks continue - if room_id in joined_room_ids or has_join: + if room_id in sync_result_builder.joined_room_ids or has_join: old_state_ids = yield self.get_state_at(room_id, since_token) old_mem_ev_id = old_state_ids.get((EventTypes.Member, user_id), None) old_mem_ev = None @@ -1079,7 +1078,7 @@ class SyncHandler(object): newly_joined_rooms.append(room_id) # If user is in the room then we don't need to do the invite/leave checks - if room_id in joined_room_ids: + if room_id in sync_result_builder.joined_room_ids: continue if not non_joins: @@ -1146,7 +1145,7 @@ class SyncHandler(object): # Get all events for rooms we're currently joined to. room_to_events = yield self.store.get_room_events_stream_for_rooms( - room_ids=joined_room_ids, + room_ids=sync_result_builder.joined_room_ids, from_key=since_token.room_key, to_key=now_token.room_key, limit=timeline_limit + 1, @@ -1154,7 +1153,7 @@ class SyncHandler(object): # We loop through all room ids, even if there are no new events, in case # there are non room events taht we need to notify about. - for room_id in joined_room_ids: + for room_id in sync_result_builder.joined_room_ids: room_entry = room_to_events.get(room_id, None) if room_entry: @@ -1362,6 +1361,54 @@ class SyncHandler(object): else: raise Exception("Unrecognized rtype: %r", room_builder.rtype) + @defer.inlineCallbacks + def get_rooms_for_user_at(self, user_id, stream_ordering): + """Get set of joined rooms for a user at the given stream ordering. + + The stream ordering *must* be recent, otherwise this may throw an + exception if older than a month. (This function is called with the + current token, which should be perfectly fine). + + Args: + user_id (str) + stream_ordering (int) + + ReturnValue: + Deferred[frozenset[str]]: Set of room_ids the user is in at given + stream_ordering. + """ + joined_rooms = yield self.store.get_rooms_for_user_with_stream_ordering( + user_id, + ) + + joined_room_ids = set() + + # We need to check that the stream ordering of the join for each room + # is before the stream_ordering asked for. This might not be the case + # if the user joins a room between us getting the current token and + # calling `get_rooms_for_user_with_stream_ordering`. + # If the membership's stream ordering is after the given stream + # ordering, we need to go and work out if the user was in the room + # before. + for room_id, membership_stream_ordering in joined_rooms: + if membership_stream_ordering <= stream_ordering: + joined_room_ids.add(room_id) + continue + + logger.info("User joined room after current token: %s", room_id) + + extrems = yield self.store.get_forward_extremeties_for_room( + room_id, stream_ordering, + ) + users_in_room = yield self.state.get_current_user_in_room( + room_id, extrems, + ) + if user_id in users_in_room: + joined_room_ids.add(room_id) + + joined_room_ids = frozenset(joined_room_ids) + defer.returnValue(joined_room_ids) + def _action_has_highlight(actions): for action in actions: @@ -1411,7 +1458,8 @@ def _calculate_state(timeline_contains, timeline_start, previous, current): class SyncResultBuilder(object): "Used to help build up a new SyncResult for a user" - def __init__(self, sync_config, full_state, since_token, now_token): + def __init__(self, sync_config, full_state, since_token, now_token, + joined_room_ids): """ Args: sync_config(SyncConfig) @@ -1423,6 +1471,7 @@ class SyncResultBuilder(object): self.full_state = full_state self.since_token = since_token self.now_token = now_token + self.joined_room_ids = joined_room_ids self.presence = [] self.account_data = [] diff --git a/synapse/handlers/typing.py b/synapse/handlers/typing.py index 82dedbbc99..77c0cf146f 100644 --- a/synapse/handlers/typing.py +++ b/synapse/handlers/typing.py @@ -56,7 +56,7 @@ class TypingHandler(object): self.federation = hs.get_federation_sender() - hs.get_replication_layer().register_edu_handler("m.typing", self._recv_edu) + hs.get_federation_registry().register_edu_handler("m.typing", self._recv_edu) hs.get_distributor().observe("user_left_room", self.user_left_room) diff --git a/synapse/http/client.py b/synapse/http/client.py index 4abb479ae3..f3e4973c2e 100644 --- a/synapse/http/client.py +++ b/synapse/http/client.py @@ -18,6 +18,7 @@ from OpenSSL.SSL import VERIFY_NONE from synapse.api.errors import ( CodeMessageException, MatrixCodeMessageException, SynapseError, Codes, ) +from synapse.util.caches import CACHE_SIZE_FACTOR from synapse.util.logcontext import make_deferred_yieldable from synapse.util import logcontext import synapse.metrics @@ -30,6 +31,7 @@ from twisted.internet.endpoints import HostnameEndpoint, wrapClientTLS from twisted.web.client import ( BrowserLikeRedirectAgent, ContentDecoderAgent, GzipDecoder, Agent, readBody, PartialDownloadError, + HTTPConnectionPool, ) from twisted.web.client import FileBodyProducer as TwistedFileBodyProducer from twisted.web.http import PotentialDataLoss @@ -64,13 +66,23 @@ class SimpleHttpClient(object): """ def __init__(self, hs): self.hs = hs + + pool = HTTPConnectionPool(reactor) + + # the pusher makes lots of concurrent SSL connections to sygnal, and + # tends to do so in batches, so we need to allow the pool to keep lots + # of idle connections around. + pool.maxPersistentPerHost = max((100 * CACHE_SIZE_FACTOR, 5)) + pool.cachedConnectionTimeout = 2 * 60 + # The default context factory in Twisted 14.0.0 (which we require) is # BrowserLikePolicyForHTTPS which will do regular cert validation # 'like a browser' self.agent = Agent( reactor, connectTimeout=15, - contextFactory=hs.get_http_client_context_factory() + contextFactory=hs.get_http_client_context_factory(), + pool=pool, ) self.user_agent = hs.version_string self.clock = hs.get_clock() diff --git a/synapse/http/endpoint.py b/synapse/http/endpoint.py index e2b99ef3bd..87639b9151 100644 --- a/synapse/http/endpoint.py +++ b/synapse/http/endpoint.py @@ -357,8 +357,7 @@ def _get_hosts_for_srv_record(dns_client, host): def eb(res, record_type): if res.check(DNSNameError): return [] - logger.warn("Error looking up %s for %s: %s", - record_type, host, res, res.value) + logger.warn("Error looking up %s for %s: %s", record_type, host, res) return res # no logcontexts here, so we can safely fire these off and gatherResults diff --git a/synapse/http/matrixfederationclient.py b/synapse/http/matrixfederationclient.py index 833496b72d..9145405cb0 100644 --- a/synapse/http/matrixfederationclient.py +++ b/synapse/http/matrixfederationclient.py @@ -27,7 +27,7 @@ import synapse.metrics from canonicaljson import encode_canonical_json from synapse.api.errors import ( - SynapseError, Codes, HttpResponseException, + SynapseError, Codes, HttpResponseException, FederationDeniedError, ) from signedjson.sign import sign_json @@ -123,11 +123,22 @@ class MatrixFederationHttpClient(object): Fails with ``HTTPRequestException``: if we get an HTTP response code >= 300. + Fails with ``NotRetryingDestination`` if we are not yet ready to retry this server. + + Fails with ``FederationDeniedError`` if this destination + is not on our federation whitelist + (May also fail with plenty of other Exceptions for things like DNS failures, connection failures, SSL failures.) """ + if ( + self.hs.config.federation_domain_whitelist and + destination not in self.hs.config.federation_domain_whitelist + ): + raise FederationDeniedError(destination) + limiter = yield synapse.util.retryutils.get_retry_limiter( destination, self.clock, @@ -308,6 +319,9 @@ class MatrixFederationHttpClient(object): Fails with ``NotRetryingDestination`` if we are not yet ready to retry this server. + + Fails with ``FederationDeniedError`` if this destination + is not on our federation whitelist """ if not json_data_callback: @@ -368,6 +382,9 @@ class MatrixFederationHttpClient(object): Fails with ``NotRetryingDestination`` if we are not yet ready to retry this server. + + Fails with ``FederationDeniedError`` if this destination + is not on our federation whitelist """ def body_callback(method, url_bytes, headers_dict): @@ -422,6 +439,9 @@ class MatrixFederationHttpClient(object): Fails with ``NotRetryingDestination`` if we are not yet ready to retry this server. + + Fails with ``FederationDeniedError`` if this destination + is not on our federation whitelist """ logger.debug("get_json args: %s", args) @@ -475,6 +495,9 @@ class MatrixFederationHttpClient(object): Fails with ``NotRetryingDestination`` if we are not yet ready to retry this server. + + Fails with ``FederationDeniedError`` if this destination + is not on our federation whitelist """ response = yield self._request( @@ -518,6 +541,9 @@ class MatrixFederationHttpClient(object): Fails with ``NotRetryingDestination`` if we are not yet ready to retry this server. + + Fails with ``FederationDeniedError`` if this destination + is not on our federation whitelist """ encoded_args = {} diff --git a/synapse/http/server.py b/synapse/http/server.py index f1e9002e4d..1551db239d 100644 --- a/synapse/http/server.py +++ b/synapse/http/server.py @@ -1,5 +1,6 @@ # -*- coding: utf-8 -*- # Copyright 2014-2016 OpenMarket Ltd +# Copyright 2018 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. @@ -42,36 +43,75 @@ logger = logging.getLogger(__name__) metrics = synapse.metrics.get_metrics_for(__name__) -incoming_requests_counter = metrics.register_counter( - "requests", +# total number of responses served, split by method/servlet/tag +response_count = metrics.register_counter( + "response_count", labels=["method", "servlet", "tag"], + alternative_names=( + # the following are all deprecated aliases for the same metric + metrics.name_prefix + x for x in ( + "_requests", + "_response_time:count", + "_response_ru_utime:count", + "_response_ru_stime:count", + "_response_db_txn_count:count", + "_response_db_txn_duration:count", + ) + ) ) + +requests_counter = metrics.register_counter( + "requests_received", + labels=["method", "servlet", ], +) + outgoing_responses_counter = metrics.register_counter( "responses", labels=["method", "code"], ) -response_timer = metrics.register_distribution( - "response_time", - labels=["method", "servlet", "tag"] +response_timer = metrics.register_counter( + "response_time_seconds", + labels=["method", "servlet", "tag"], + alternative_names=( + metrics.name_prefix + "_response_time:total", + ), ) -response_ru_utime = metrics.register_distribution( - "response_ru_utime", labels=["method", "servlet", "tag"] +response_ru_utime = metrics.register_counter( + "response_ru_utime_seconds", labels=["method", "servlet", "tag"], + alternative_names=( + metrics.name_prefix + "_response_ru_utime:total", + ), ) -response_ru_stime = metrics.register_distribution( - "response_ru_stime", labels=["method", "servlet", "tag"] +response_ru_stime = metrics.register_counter( + "response_ru_stime_seconds", labels=["method", "servlet", "tag"], + alternative_names=( + metrics.name_prefix + "_response_ru_stime:total", + ), ) -response_db_txn_count = metrics.register_distribution( - "response_db_txn_count", labels=["method", "servlet", "tag"] +response_db_txn_count = metrics.register_counter( + "response_db_txn_count", labels=["method", "servlet", "tag"], + alternative_names=( + metrics.name_prefix + "_response_db_txn_count:total", + ), ) -response_db_txn_duration = metrics.register_distribution( - "response_db_txn_duration", labels=["method", "servlet", "tag"] +# seconds spent waiting for db txns, excluding scheduling time, when processing +# this request +response_db_txn_duration = metrics.register_counter( + "response_db_txn_duration_seconds", labels=["method", "servlet", "tag"], + alternative_names=( + metrics.name_prefix + "_response_db_txn_duration:total", + ), ) +# seconds spent waiting for a db connection, when processing this request +response_db_sched_duration = metrics.register_counter( + "response_db_sched_duration_seconds", labels=["method", "servlet", "tag"] +) _next_request_id = 0 @@ -107,7 +147,12 @@ def wrap_request_handler(request_handler, include_metrics=False): with LoggingContext(request_id) as request_context: with Measure(self.clock, "wrapped_request_handler"): request_metrics = RequestMetrics() - request_metrics.start(self.clock, name=self.__class__.__name__) + # we start the request metrics timer here with an initial stab + # at the servlet name. For most requests that name will be + # JsonResource (or a subclass), and JsonResource._async_render + # will update it once it picks a servlet. + servlet_name = self.__class__.__name__ + request_metrics.start(self.clock, name=servlet_name) request_context.request = request_id with request.processing(): @@ -116,6 +161,7 @@ def wrap_request_handler(request_handler, include_metrics=False): if include_metrics: yield request_handler(self, request, request_metrics) else: + requests_counter.inc(request.method, servlet_name) yield request_handler(self, request) except CodeMessageException as e: code = e.code @@ -191,7 +237,7 @@ class JsonResource(HttpServer, resource.Resource): """ This implements the HttpServer interface and provides JSON support for Resources. - Register callbacks via register_path() + Register callbacks via register_paths() Callbacks can return a tuple of status code and a dict in which case the the dict will automatically be sent to the client as a JSON object. @@ -238,57 +284,62 @@ class JsonResource(HttpServer, resource.Resource): This checks if anyone has registered a callback for that method and path. """ - if request.method == "OPTIONS": - self._send_response(request, 200, {}) - return + callback, group_dict = self._get_handler_for_request(request) - # Loop through all the registered callbacks to check if the method - # and path regex match - for path_entry in self.path_regexs.get(request.method, []): - m = path_entry.pattern.match(request.path) - if not m: - continue + servlet_instance = getattr(callback, "__self__", None) + if servlet_instance is not None: + servlet_classname = servlet_instance.__class__.__name__ + else: + servlet_classname = "%r" % callback - # We found a match! Trigger callback and then return the - # returned response. We pass both the request and any - # matched groups from the regex to the callback. + request_metrics.name = servlet_classname + requests_counter.inc(request.method, servlet_classname) - callback = path_entry.callback + # Now trigger the callback. If it returns a response, we send it + # here. If it throws an exception, that is handled by the wrapper + # installed by @request_handler. - kwargs = intern_dict({ - name: urllib.unquote(value).decode("UTF-8") if value else value - for name, value in m.groupdict().items() - }) + kwargs = intern_dict({ + name: urllib.unquote(value).decode("UTF-8") if value else value + for name, value in group_dict.items() + }) - callback_return = yield callback(request, **kwargs) - if callback_return is not None: - code, response = callback_return - self._send_response(request, code, response) + callback_return = yield callback(request, **kwargs) + if callback_return is not None: + code, response = callback_return + self._send_response(request, code, response) - servlet_instance = getattr(callback, "__self__", None) - if servlet_instance is not None: - servlet_classname = servlet_instance.__class__.__name__ - else: - servlet_classname = "%r" % callback + def _get_handler_for_request(self, request): + """Finds a callback method to handle the given request + + Args: + request (twisted.web.http.Request): - request_metrics.name = servlet_classname + Returns: + Tuple[Callable, dict[str, str]]: callback method, and the dict + mapping keys to path components as specified in the handler's + path match regexp. - return + The callback will normally be a method registered via + register_paths, so will return (possibly via Deferred) either + None, or a tuple of (http code, response body). + """ + if request.method == "OPTIONS": + return _options_handler, {} + + # Loop through all the registered callbacks to check if the method + # and path regex match + for path_entry in self.path_regexs.get(request.method, []): + m = path_entry.pattern.match(request.path) + if m: + # We found a match! + return path_entry.callback, m.groupdict() # Huh. No one wanted to handle that? Fiiiiiine. Send 400. - raise UnrecognizedRequestError() + return _unrecognised_request_handler, {} def _send_response(self, request, code, response_json_object, response_code_message=None): - # could alternatively use request.notifyFinish() and flip a flag when - # the Deferred fires, but since the flag is RIGHT THERE it seems like - # a waste. - if request._disconnected: - logger.warn( - "Not sending response to request %s, already disconnected.", - request) - return - outgoing_responses_counter.inc(request.method, str(code)) # TODO: Only enable CORS for the requests that need it. @@ -302,6 +353,34 @@ class JsonResource(HttpServer, resource.Resource): ) +def _options_handler(request): + """Request handler for OPTIONS requests + + This is a request handler suitable for return from + _get_handler_for_request. It returns a 200 and an empty body. + + Args: + request (twisted.web.http.Request): + + Returns: + Tuple[int, dict]: http code, response body. + """ + return 200, {} + + +def _unrecognised_request_handler(request): + """Request handler for unrecognised requests + + This is a request handler suitable for return from + _get_handler_for_request. It actually just raises an + UnrecognizedRequestError. + + Args: + request (twisted.web.http.Request): + """ + raise UnrecognizedRequestError() + + class RequestMetrics(object): def start(self, clock, name): self.start = clock.time_msec() @@ -322,7 +401,7 @@ class RequestMetrics(object): ) return - incoming_requests_counter.inc(request.method, self.name, tag) + response_count.inc(request.method, self.name, tag) response_timer.inc_by( clock.time_msec() - self.start, request.method, @@ -341,7 +420,10 @@ class RequestMetrics(object): context.db_txn_count, request.method, self.name, tag ) response_db_txn_duration.inc_by( - context.db_txn_duration, request.method, self.name, tag + context.db_txn_duration_ms / 1000., request.method, self.name, tag + ) + response_db_sched_duration.inc_by( + context.db_sched_duration_ms / 1000., request.method, self.name, tag ) @@ -364,6 +446,15 @@ class RootRedirect(resource.Resource): def respond_with_json(request, code, json_object, send_cors=False, response_code_message=None, pretty_print=False, version_string="", canonical_json=True): + # could alternatively use request.notifyFinish() and flip a flag when + # the Deferred fires, but since the flag is RIGHT THERE it seems like + # a waste. + if request._disconnected: + logger.warn( + "Not sending response to request %s, already disconnected.", + request) + return + if pretty_print: json_bytes = encode_pretty_printed_json(json_object) + "\n" else: diff --git a/synapse/http/servlet.py b/synapse/http/servlet.py index 71420e54db..ef8e62901b 100644 --- a/synapse/http/servlet.py +++ b/synapse/http/servlet.py @@ -148,11 +148,13 @@ def parse_string_from_args(args, name, default=None, required=False, return default -def parse_json_value_from_request(request): +def parse_json_value_from_request(request, allow_empty_body=False): """Parse a JSON value from the body of a twisted HTTP request. Args: request: the twisted HTTP request. + allow_empty_body (bool): if True, an empty body will be accepted and + turned into None Returns: The JSON value. @@ -165,6 +167,9 @@ def parse_json_value_from_request(request): except Exception: raise SynapseError(400, "Error reading JSON content.") + if not content_bytes and allow_empty_body: + return None + try: content = simplejson.loads(content_bytes) except Exception as e: @@ -174,17 +179,24 @@ def parse_json_value_from_request(request): return content -def parse_json_object_from_request(request): +def parse_json_object_from_request(request, allow_empty_body=False): """Parse a JSON object from the body of a twisted HTTP request. Args: request: the twisted HTTP request. + allow_empty_body (bool): if True, an empty body will be accepted and + turned into an empty dict. Raises: SynapseError if the request body couldn't be decoded as JSON or if it wasn't a JSON object. """ - content = parse_json_value_from_request(request) + content = parse_json_value_from_request( + request, allow_empty_body=allow_empty_body, + ) + + if allow_empty_body and content is None: + return {} if type(content) != dict: message = "Content must be a JSON object." diff --git a/synapse/http/site.py b/synapse/http/site.py index cd1492b1c3..e422c8dfae 100644 --- a/synapse/http/site.py +++ b/synapse/http/site.py @@ -66,14 +66,15 @@ class SynapseRequest(Request): context = LoggingContext.current_context() ru_utime, ru_stime = context.get_resource_usage() db_txn_count = context.db_txn_count - db_txn_duration = context.db_txn_duration + db_txn_duration_ms = context.db_txn_duration_ms + db_sched_duration_ms = context.db_sched_duration_ms except Exception: ru_utime, ru_stime = (0, 0) - db_txn_count, db_txn_duration = (0, 0) + db_txn_count, db_txn_duration_ms = (0, 0) self.site.access_logger.info( "%s - %s - {%s}" - " Processed request: %dms (%dms, %dms) (%dms/%d)" + " Processed request: %dms (%dms, %dms) (%dms/%dms/%d)" " %sB %s \"%s %s %s\" \"%s\"", self.getClientIP(), self.site.site_tag, @@ -81,7 +82,8 @@ class SynapseRequest(Request): int(time.time() * 1000) - self.start_time, int(ru_utime * 1000), int(ru_stime * 1000), - int(db_txn_duration * 1000), + db_sched_duration_ms, + db_txn_duration_ms, int(db_txn_count), self.sentLength, self.code, diff --git a/synapse/metrics/__init__.py b/synapse/metrics/__init__.py index 2265e6e8d6..50d99d7a5c 100644 --- a/synapse/metrics/__init__.py +++ b/synapse/metrics/__init__.py @@ -57,15 +57,31 @@ class Metrics(object): return metric def register_counter(self, *args, **kwargs): + """ + Returns: + CounterMetric + """ return self._register(CounterMetric, *args, **kwargs) def register_callback(self, *args, **kwargs): + """ + Returns: + CallbackMetric + """ return self._register(CallbackMetric, *args, **kwargs) def register_distribution(self, *args, **kwargs): + """ + Returns: + DistributionMetric + """ return self._register(DistributionMetric, *args, **kwargs) def register_cache(self, *args, **kwargs): + """ + Returns: + CacheMetric + """ return self._register(CacheMetric, *args, **kwargs) @@ -146,10 +162,15 @@ def runUntilCurrentTimer(func): num_pending += 1 num_pending += len(reactor.threadCallQueue) - start = time.time() * 1000 ret = func(*args, **kwargs) end = time.time() * 1000 + + # record the amount of wallclock time spent running pending calls. + # This is a proxy for the actual amount of time between reactor polls, + # since about 25% of time is actually spent running things triggered by + # I/O events, but that is harder to capture without rewriting half the + # reactor. tick_time.inc_by(end - start) pending_calls_metric.inc_by(num_pending) diff --git a/synapse/metrics/metric.py b/synapse/metrics/metric.py index e87b2b80a7..ff5aa8c0e1 100644 --- a/synapse/metrics/metric.py +++ b/synapse/metrics/metric.py @@ -15,18 +15,38 @@ from itertools import chain +import logging +logger = logging.getLogger(__name__) -# TODO(paul): I can't believe Python doesn't have one of these -def map_concat(func, items): - # flatten a list-of-lists - return list(chain.from_iterable(map(func, items))) + +def flatten(items): + """Flatten a list of lists + + Args: + items: iterable[iterable[X]] + + Returns: + list[X]: flattened list + """ + return list(chain.from_iterable(items)) class BaseMetric(object): + """Base class for metrics which report a single value per label set + """ - def __init__(self, name, labels=[]): - self.name = name + def __init__(self, name, labels=[], alternative_names=[]): + """ + Args: + name (str): principal name for this metric + labels (list(str)): names of the labels which will be reported + for this metric + alternative_names (iterable(str)): list of alternative names for + this metric. This can be useful to provide a migration path + when renaming metrics. + """ + self._names = [name] + list(alternative_names) self.labels = labels # OK not to clone as we never write it def dimension(self): @@ -36,7 +56,7 @@ class BaseMetric(object): return not len(self.labels) def _render_labelvalue(self, value): - # TODO: some kind of value escape + # TODO: escape backslashes, quotes and newlines return '"%s"' % (value) def _render_key(self, values): @@ -47,19 +67,60 @@ class BaseMetric(object): for k, v in zip(self.labels, values)]) ) + def _render_for_labels(self, label_values, value): + """Render this metric for a single set of labels + + Args: + label_values (list[str]): values for each of the labels + value: value of the metric at with these labels + + Returns: + iterable[str]: rendered metric + """ + rendered_labels = self._render_key(label_values) + return ( + "%s%s %.12g" % (name, rendered_labels, value) + for name in self._names + ) + + def render(self): + """Render this metric + + Each metric is rendered as: + + name{label1="val1",label2="val2"} value + + https://prometheus.io/docs/instrumenting/exposition_formats/#text-format-details + + Returns: + iterable[str]: rendered metrics + """ + raise NotImplementedError() + class CounterMetric(BaseMetric): """The simplest kind of metric; one that stores a monotonically-increasing - integer that counts events.""" + value that counts events or running totals. + + Example use cases for Counters: + - Number of requests processed + - Number of items that were inserted into a queue + - Total amount of data that a system has processed + Counters can only go up (and be reset when the process restarts). + """ def __init__(self, *args, **kwargs): super(CounterMetric, self).__init__(*args, **kwargs) + # dict[list[str]]: value for each set of label values. the keys are the + # label values, in the same order as the labels in self.labels. + # + # (if the metric is a scalar, the (single) key is the empty list). self.counts = {} # Scalar metrics are never empty if self.is_scalar(): - self.counts[()] = 0 + self.counts[()] = 0. def inc_by(self, incr, *values): if len(values) != self.dimension(): @@ -77,11 +138,11 @@ class CounterMetric(BaseMetric): def inc(self, *values): self.inc_by(1, *values) - def render_item(self, k): - return ["%s%s %d" % (self.name, self._render_key(k), self.counts[k])] - def render(self): - return map_concat(self.render_item, sorted(self.counts.keys())) + return flatten( + self._render_for_labels(k, self.counts[k]) + for k in sorted(self.counts.keys()) + ) class CallbackMetric(BaseMetric): @@ -95,13 +156,19 @@ class CallbackMetric(BaseMetric): self.callback = callback def render(self): - value = self.callback() + try: + value = self.callback() + except Exception: + logger.exception("Failed to render %s", self.name) + return ["# FAILED to render " + self.name] if self.is_scalar(): - return ["%s %.12g" % (self.name, value)] + return list(self._render_for_labels([], value)) - return ["%s%s %.12g" % (self.name, self._render_key(k), value[k]) - for k in sorted(value.keys())] + return flatten( + self._render_for_labels(k, value[k]) + for k in sorted(value.keys()) + ) class DistributionMetric(object): @@ -126,7 +193,9 @@ class DistributionMetric(object): class CacheMetric(object): - __slots__ = ("name", "cache_name", "hits", "misses", "size_callback") + __slots__ = ( + "name", "cache_name", "hits", "misses", "evicted_size", "size_callback", + ) def __init__(self, name, size_callback, cache_name): self.name = name @@ -134,6 +203,7 @@ class CacheMetric(object): self.hits = 0 self.misses = 0 + self.evicted_size = 0 self.size_callback = size_callback @@ -143,6 +213,9 @@ class CacheMetric(object): def inc_misses(self): self.misses += 1 + def inc_evictions(self, size=1): + self.evicted_size += size + def render(self): size = self.size_callback() hits = self.hits @@ -152,6 +225,9 @@ class CacheMetric(object): """%s:hits{name="%s"} %d""" % (self.name, self.cache_name, hits), """%s:total{name="%s"} %d""" % (self.name, self.cache_name, total), """%s:size{name="%s"} %d""" % (self.name, self.cache_name, size), + """%s:evicted_size{name="%s"} %d""" % ( + self.name, self.cache_name, self.evicted_size + ), ] diff --git a/synapse/push/action_generator.py b/synapse/push/action_generator.py index fe09d50d55..8f619a7a1b 100644 --- a/synapse/push/action_generator.py +++ b/synapse/push/action_generator.py @@ -40,10 +40,6 @@ class ActionGenerator(object): @defer.inlineCallbacks def handle_push_actions_for_event(self, event, context): with Measure(self.clock, "action_for_event_by_user"): - actions_by_user = yield self.bulk_evaluator.action_for_event_by_user( + yield self.bulk_evaluator.action_for_event_by_user( event, context ) - - context.push_actions = [ - (uid, actions) for uid, actions in actions_by_user.iteritems() - ] diff --git a/synapse/push/bulk_push_rule_evaluator.py b/synapse/push/bulk_push_rule_evaluator.py index 425a017bdf..7c680659b6 100644 --- a/synapse/push/bulk_push_rule_evaluator.py +++ b/synapse/push/bulk_push_rule_evaluator.py @@ -137,11 +137,11 @@ class BulkPushRuleEvaluator(object): @defer.inlineCallbacks def action_for_event_by_user(self, event, context): - """Given an event and context, evaluate the push rules and return - the results + """Given an event and context, evaluate the push rules and insert the + results into the event_push_actions_staging table. Returns: - dict of user_id -> action + Deferred """ rules_by_user = yield self._get_rules_for_event(event, context) actions_by_user = {} @@ -190,9 +190,16 @@ class BulkPushRuleEvaluator(object): if matches: actions = [x for x in rule['actions'] if x != 'dont_notify'] if actions and 'notify' in actions: + # Push rules say we should notify the user of this event actions_by_user[uid] = actions break - defer.returnValue(actions_by_user) + + # Mark in the DB staging area the push actions for users who should be + # notified for this event. (This will then get handled when we persist + # the event) + yield self.store.add_push_actions_to_staging( + event.event_id, actions_by_user, + ) def _condition_checker(evaluator, conditions, uid, display_name, cache): diff --git a/synapse/push/httppusher.py b/synapse/push/httppusher.py index c16f61452c..2cbac571b8 100644 --- a/synapse/push/httppusher.py +++ b/synapse/push/httppusher.py @@ -13,21 +13,30 @@ # 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 synapse.push import PusherConfigException +import logging from twisted.internet import defer, reactor from twisted.internet.error import AlreadyCalled, AlreadyCancelled -import logging import push_rule_evaluator import push_tools - +import synapse +from synapse.push import PusherConfigException from synapse.util.logcontext import LoggingContext from synapse.util.metrics import Measure logger = logging.getLogger(__name__) +metrics = synapse.metrics.get_metrics_for(__name__) + +http_push_processed_counter = metrics.register_counter( + "http_pushes_processed", +) + +http_push_failed_counter = metrics.register_counter( + "http_pushes_failed", +) + class HttpPusher(object): INITIAL_BACKOFF_SEC = 1 # in seconds because that's what Twisted takes @@ -152,9 +161,16 @@ class HttpPusher(object): self.user_id, self.last_stream_ordering, self.max_stream_ordering ) + logger.info( + "Processing %i unprocessed push actions for %s starting at " + "stream_ordering %s", + len(unprocessed), self.name, self.last_stream_ordering, + ) + for push_action in unprocessed: processed = yield self._process_one(push_action) if processed: + http_push_processed_counter.inc() self.backoff_delay = HttpPusher.INITIAL_BACKOFF_SEC self.last_stream_ordering = push_action['stream_ordering'] yield self.store.update_pusher_last_stream_ordering_and_success( @@ -169,6 +185,7 @@ class HttpPusher(object): self.failing_since ) else: + http_push_failed_counter.inc() if not self.failing_since: self.failing_since = self.clock.time_msec() yield self.store.update_pusher_failing_since( @@ -316,7 +333,10 @@ class HttpPusher(object): try: resp = yield self.http_client.post_json_get_json(self.url, notification_dict) except Exception: - logger.warn("Failed to push %s ", self.url) + logger.warn( + "Failed to push event %s to %s", + event.event_id, self.name, exc_info=True, + ) defer.returnValue(False) rejected = [] if 'rejected' in resp: @@ -325,7 +345,7 @@ class HttpPusher(object): @defer.inlineCallbacks def _send_badge(self, badge): - logger.info("Sending updated badge count %d to %r", badge, self.user_id) + logger.info("Sending updated badge count %d to %s", badge, self.name) d = { 'notification': { 'id': '', @@ -347,7 +367,10 @@ class HttpPusher(object): try: resp = yield self.http_client.post_json_get_json(self.url, d) except Exception: - logger.exception("Failed to push %s ", self.url) + logger.warn( + "Failed to send badge count to %s", + self.name, exc_info=True, + ) defer.returnValue(False) rejected = [] if 'rejected' in resp: diff --git a/synapse/python_dependencies.py b/synapse/python_dependencies.py index 7052333c19..91179ce532 100644 --- a/synapse/python_dependencies.py +++ b/synapse/python_dependencies.py @@ -24,19 +24,19 @@ REQUIREMENTS = { "unpaddedbase64>=1.1.0": ["unpaddedbase64>=1.1.0"], "canonicaljson>=1.0.0": ["canonicaljson>=1.0.0"], "signedjson>=1.0.0": ["signedjson>=1.0.0"], - "pynacl==0.3.0": ["nacl==0.3.0", "nacl.bindings"], + "pynacl>=1.2.1": ["nacl>=1.2.1", "nacl.bindings"], "service_identity>=1.0.0": ["service_identity>=1.0.0"], "Twisted>=16.0.0": ["twisted>=16.0.0"], "pyopenssl>=0.14": ["OpenSSL>=0.14"], "pyyaml": ["yaml"], "pyasn1": ["pyasn1"], "daemonize": ["daemonize"], - "bcrypt": ["bcrypt"], + "bcrypt": ["bcrypt>=3.1.0"], "pillow": ["PIL"], "pydenticon": ["pydenticon"], "ujson": ["ujson"], "blist": ["blist"], - "pysaml2>=3.0.0,<4.0.0": ["saml2>=3.0.0,<4.0.0"], + "pysaml2>=3.0.0": ["saml2>=3.0.0"], "pymacaroons-pynacl": ["pymacaroons"], "msgpack-python>=0.3.0": ["msgpack"], "phonenumbers>=8.2.0": ["phonenumbers"], diff --git a/synapse/replication/http/__init__.py b/synapse/replication/http/__init__.py new file mode 100644 index 0000000000..1d7a607529 --- /dev/null +++ b/synapse/replication/http/__init__.py @@ -0,0 +1,30 @@ +# -*- coding: utf-8 -*- +# Copyright 2018 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. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +from synapse.http.server import JsonResource +from synapse.replication.http import membership, send_event + + +REPLICATION_PREFIX = "/_synapse/replication" + + +class ReplicationRestResource(JsonResource): + def __init__(self, hs): + JsonResource.__init__(self, hs, canonical_json=False) + self.register_servlets(hs) + + def register_servlets(self, hs): + send_event.register_servlets(hs, self) + membership.register_servlets(hs, self) diff --git a/synapse/replication/http/membership.py b/synapse/replication/http/membership.py new file mode 100644 index 0000000000..e66c4e881f --- /dev/null +++ b/synapse/replication/http/membership.py @@ -0,0 +1,334 @@ +# -*- coding: utf-8 -*- +# Copyright 2018 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. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +import logging +import re + +from twisted.internet import defer + +from synapse.api.errors import SynapseError, MatrixCodeMessageException +from synapse.http.servlet import RestServlet, parse_json_object_from_request +from synapse.types import Requester, UserID +from synapse.util.distributor import user_left_room, user_joined_room + +logger = logging.getLogger(__name__) + + +@defer.inlineCallbacks +def remote_join(client, host, port, requester, remote_room_hosts, + room_id, user_id, content): + """Ask the master to do a remote join for the given user to the given room + + Args: + client (SimpleHttpClient) + host (str): host of master + port (int): port on master listening for HTTP replication + requester (Requester) + remote_room_hosts (list[str]): Servers to try and join via + room_id (str) + user_id (str) + content (dict): The event content to use for the join event + + Returns: + Deferred + """ + uri = "http://%s:%s/_synapse/replication/remote_join" % (host, port) + + payload = { + "requester": requester.serialize(), + "remote_room_hosts": remote_room_hosts, + "room_id": room_id, + "user_id": user_id, + "content": content, + } + + try: + result = yield client.post_json_get_json(uri, payload) + except MatrixCodeMessageException as e: + # We convert to SynapseError as we know that it was a SynapseError + # on the master process that we should send to the client. (And + # importantly, not stack traces everywhere) + raise SynapseError(e.code, e.msg, e.errcode) + defer.returnValue(result) + + +@defer.inlineCallbacks +def remote_reject_invite(client, host, port, requester, remote_room_hosts, + room_id, user_id): + """Ask master to reject the invite for the user and room. + + Args: + client (SimpleHttpClient) + host (str): host of master + port (int): port on master listening for HTTP replication + requester (Requester) + remote_room_hosts (list[str]): Servers to try and reject via + room_id (str) + user_id (str) + + Returns: + Deferred + """ + uri = "http://%s:%s/_synapse/replication/remote_reject_invite" % (host, port) + + payload = { + "requester": requester.serialize(), + "remote_room_hosts": remote_room_hosts, + "room_id": room_id, + "user_id": user_id, + } + + try: + result = yield client.post_json_get_json(uri, payload) + except MatrixCodeMessageException as e: + # We convert to SynapseError as we know that it was a SynapseError + # on the master process that we should send to the client. (And + # importantly, not stack traces everywhere) + raise SynapseError(e.code, e.msg, e.errcode) + defer.returnValue(result) + + +@defer.inlineCallbacks +def get_or_register_3pid_guest(client, host, port, requester, + medium, address, inviter_user_id): + """Ask the master to get/create a guest account for given 3PID. + + Args: + client (SimpleHttpClient) + host (str): host of master + port (int): port on master listening for HTTP replication + requester (Requester) + medium (str) + address (str) + inviter_user_id (str): The user ID who is trying to invite the + 3PID + + Returns: + Deferred[(str, str)]: A 2-tuple of `(user_id, access_token)` of the + 3PID guest account. + """ + + uri = "http://%s:%s/_synapse/replication/get_or_register_3pid_guest" % (host, port) + + payload = { + "requester": requester.serialize(), + "medium": medium, + "address": address, + "inviter_user_id": inviter_user_id, + } + + try: + result = yield client.post_json_get_json(uri, payload) + except MatrixCodeMessageException as e: + # We convert to SynapseError as we know that it was a SynapseError + # on the master process that we should send to the client. (And + # importantly, not stack traces everywhere) + raise SynapseError(e.code, e.msg, e.errcode) + defer.returnValue(result) + + +@defer.inlineCallbacks +def notify_user_membership_change(client, host, port, user_id, room_id, change): + """Notify master that a user has joined or left the room + + Args: + client (SimpleHttpClient) + host (str): host of master + port (int): port on master listening for HTTP replication. + user_id (str) + room_id (str) + change (str): Either "join" or "left" + + Returns: + Deferred + """ + assert change in ("joined", "left") + + uri = "http://%s:%s/_synapse/replication/user_%s_room" % (host, port, change) + + payload = { + "user_id": user_id, + "room_id": room_id, + } + + try: + result = yield client.post_json_get_json(uri, payload) + except MatrixCodeMessageException as e: + # We convert to SynapseError as we know that it was a SynapseError + # on the master process that we should send to the client. (And + # importantly, not stack traces everywhere) + raise SynapseError(e.code, e.msg, e.errcode) + defer.returnValue(result) + + +class ReplicationRemoteJoinRestServlet(RestServlet): + PATTERNS = [re.compile("^/_synapse/replication/remote_join$")] + + def __init__(self, hs): + super(ReplicationRemoteJoinRestServlet, self).__init__() + + self.federation_handler = hs.get_handlers().federation_handler + self.store = hs.get_datastore() + self.clock = hs.get_clock() + + @defer.inlineCallbacks + def on_POST(self, request): + content = parse_json_object_from_request(request) + + remote_room_hosts = content["remote_room_hosts"] + room_id = content["room_id"] + user_id = content["user_id"] + event_content = content["content"] + + requester = Requester.deserialize(self.store, content["requester"]) + + if requester.user: + request.authenticated_entity = requester.user.to_string() + + logger.info( + "remote_join: %s into room: %s", + user_id, room_id, + ) + + yield self.federation_handler.do_invite_join( + remote_room_hosts, + room_id, + user_id, + event_content, + ) + + defer.returnValue((200, {})) + + +class ReplicationRemoteRejectInviteRestServlet(RestServlet): + PATTERNS = [re.compile("^/_synapse/replication/remote_reject_invite$")] + + def __init__(self, hs): + super(ReplicationRemoteRejectInviteRestServlet, self).__init__() + + self.federation_handler = hs.get_handlers().federation_handler + self.store = hs.get_datastore() + self.clock = hs.get_clock() + + @defer.inlineCallbacks + def on_POST(self, request): + content = parse_json_object_from_request(request) + + remote_room_hosts = content["remote_room_hosts"] + room_id = content["room_id"] + user_id = content["user_id"] + + requester = Requester.deserialize(self.store, content["requester"]) + + if requester.user: + request.authenticated_entity = requester.user.to_string() + + logger.info( + "remote_reject_invite: %s out of room: %s", + user_id, room_id, + ) + + try: + event = yield self.federation_handler.do_remotely_reject_invite( + remote_room_hosts, + room_id, + user_id, + ) + ret = event.get_pdu_json() + except Exception as e: + # if we were unable to reject the exception, just mark + # it as rejected on our end and plough ahead. + # + # The 'except' clause is very broad, but we need to + # capture everything from DNS failures upwards + # + logger.warn("Failed to reject invite: %s", e) + + yield self.store.locally_reject_invite( + user_id, room_id + ) + ret = {} + + defer.returnValue((200, ret)) + + +class ReplicationRegister3PIDGuestRestServlet(RestServlet): + PATTERNS = [re.compile("^/_synapse/replication/get_or_register_3pid_guest$")] + + def __init__(self, hs): + super(ReplicationRegister3PIDGuestRestServlet, self).__init__() + + self.registeration_handler = hs.get_handlers().registration_handler + self.store = hs.get_datastore() + self.clock = hs.get_clock() + + @defer.inlineCallbacks + def on_POST(self, request): + content = parse_json_object_from_request(request) + + medium = content["medium"] + address = content["address"] + inviter_user_id = content["inviter_user_id"] + + requester = Requester.deserialize(self.store, content["requester"]) + + if requester.user: + request.authenticated_entity = requester.user.to_string() + + logger.info("get_or_register_3pid_guest: %r", content) + + ret = yield self.registeration_handler.get_or_register_3pid_guest( + medium, address, inviter_user_id, + ) + + defer.returnValue((200, ret)) + + +class ReplicationUserJoinedLeftRoomRestServlet(RestServlet): + PATTERNS = [re.compile("^/_synapse/replication/user_(?P<change>joined|left)_room$")] + + def __init__(self, hs): + super(ReplicationUserJoinedLeftRoomRestServlet, self).__init__() + + self.registeration_handler = hs.get_handlers().registration_handler + self.store = hs.get_datastore() + self.clock = hs.get_clock() + self.distributor = hs.get_distributor() + + def on_POST(self, request, change): + content = parse_json_object_from_request(request) + + user_id = content["user_id"] + room_id = content["room_id"] + + logger.info("user membership change: %s in %s", user_id, room_id) + + user = UserID.from_string(user_id) + + if change == "joined": + user_joined_room(self.distributor, user, room_id) + elif change == "left": + user_left_room(self.distributor, user, room_id) + else: + raise Exception("Unrecognized change: %r", change) + + return (200, {}) + + +def register_servlets(hs, http_server): + ReplicationRemoteJoinRestServlet(hs).register(http_server) + ReplicationRemoteRejectInviteRestServlet(hs).register(http_server) + ReplicationRegister3PIDGuestRestServlet(hs).register(http_server) + ReplicationUserJoinedLeftRoomRestServlet(hs).register(http_server) diff --git a/synapse/replication/http/send_event.py b/synapse/replication/http/send_event.py new file mode 100644 index 0000000000..bbe2f967b7 --- /dev/null +++ b/synapse/replication/http/send_event.py @@ -0,0 +1,166 @@ +# -*- coding: utf-8 -*- +# Copyright 2018 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. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +from twisted.internet import defer + +from synapse.api.errors import ( + SynapseError, MatrixCodeMessageException, CodeMessageException, +) +from synapse.events import FrozenEvent +from synapse.events.snapshot import EventContext +from synapse.http.servlet import RestServlet, parse_json_object_from_request +from synapse.util.async import sleep +from synapse.util.caches.response_cache import ResponseCache +from synapse.util.logcontext import make_deferred_yieldable, preserve_fn +from synapse.util.metrics import Measure +from synapse.types import Requester, UserID + +import logging +import re + +logger = logging.getLogger(__name__) + + +@defer.inlineCallbacks +def send_event_to_master(client, host, port, requester, event, context, + ratelimit, extra_users): + """Send event to be handled on the master + + Args: + client (SimpleHttpClient) + host (str): host of master + port (int): port on master listening for HTTP replication + requester (Requester) + event (FrozenEvent) + context (EventContext) + ratelimit (bool) + extra_users (list(UserID)): Any extra users to notify about event + """ + uri = "http://%s:%s/_synapse/replication/send_event/%s" % ( + host, port, event.event_id, + ) + + payload = { + "event": event.get_pdu_json(), + "internal_metadata": event.internal_metadata.get_dict(), + "rejected_reason": event.rejected_reason, + "context": context.serialize(event), + "requester": requester.serialize(), + "ratelimit": ratelimit, + "extra_users": [u.to_string() for u in extra_users], + } + + try: + # We keep retrying the same request for timeouts. This is so that we + # have a good idea that the request has either succeeded or failed on + # the master, and so whether we should clean up or not. + while True: + try: + result = yield client.put_json(uri, payload) + break + except CodeMessageException as e: + if e.code != 504: + raise + + logger.warn("send_event request timed out") + + # If we timed out we probably don't need to worry about backing + # off too much, but lets just wait a little anyway. + yield sleep(1) + except MatrixCodeMessageException as e: + # We convert to SynapseError as we know that it was a SynapseError + # on the master process that we should send to the client. (And + # importantly, not stack traces everywhere) + raise SynapseError(e.code, e.msg, e.errcode) + defer.returnValue(result) + + +class ReplicationSendEventRestServlet(RestServlet): + """Handles events newly created on workers, including persisting and + notifying. + + The API looks like: + + POST /_synapse/replication/send_event/:event_id + + { + "event": { .. serialized event .. }, + "internal_metadata": { .. serialized internal_metadata .. }, + "rejected_reason": .., // The event.rejected_reason field + "context": { .. serialized event context .. }, + "requester": { .. serialized requester .. }, + "ratelimit": true, + "extra_users": [], + } + """ + PATTERNS = [re.compile("^/_synapse/replication/send_event/(?P<event_id>[^/]+)$")] + + def __init__(self, hs): + super(ReplicationSendEventRestServlet, self).__init__() + + self.event_creation_handler = hs.get_event_creation_handler() + self.store = hs.get_datastore() + self.clock = hs.get_clock() + + # The responses are tiny, so we may as well cache them for a while + self.response_cache = ResponseCache(hs, timeout_ms=30 * 60 * 1000) + + def on_PUT(self, request, event_id): + result = self.response_cache.get(event_id) + if not result: + result = self.response_cache.set( + event_id, + self._handle_request(request) + ) + else: + logger.warn("Returning cached response") + return make_deferred_yieldable(result) + + @preserve_fn + @defer.inlineCallbacks + def _handle_request(self, request): + with Measure(self.clock, "repl_send_event_parse"): + content = parse_json_object_from_request(request) + + event_dict = content["event"] + internal_metadata = content["internal_metadata"] + rejected_reason = content["rejected_reason"] + event = FrozenEvent(event_dict, internal_metadata, rejected_reason) + + requester = Requester.deserialize(self.store, content["requester"]) + context = yield EventContext.deserialize(self.store, content["context"]) + + ratelimit = content["ratelimit"] + extra_users = [UserID.from_string(u) for u in content["extra_users"]] + + if requester.user: + request.authenticated_entity = requester.user.to_string() + + logger.info( + "Got event to send with ID: %s into room: %s", + event.event_id, event.room_id, + ) + + yield self.event_creation_handler.persist_and_notify_client_event( + requester, event, context, + ratelimit=ratelimit, + extra_users=extra_users, + ) + + defer.returnValue((200, {})) + + +def register_servlets(hs, http_server): + ReplicationSendEventRestServlet(hs).register(http_server) diff --git a/synapse/replication/slave/storage/account_data.py b/synapse/replication/slave/storage/account_data.py index efbd87918e..d9ba6d69b1 100644 --- a/synapse/replication/slave/storage/account_data.py +++ b/synapse/replication/slave/storage/account_data.py @@ -1,5 +1,6 @@ # -*- coding: utf-8 -*- # Copyright 2016 OpenMarket Ltd +# Copyright 2018 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. @@ -13,50 +14,20 @@ # See the License for the specific language governing permissions and # limitations under the License. -from ._base import BaseSlavedStore -from ._slaved_id_tracker import SlavedIdTracker -from synapse.storage import DataStore -from synapse.storage.account_data import AccountDataStore -from synapse.storage.tags import TagsStore -from synapse.util.caches.stream_change_cache import StreamChangeCache +from synapse.replication.slave.storage._base import BaseSlavedStore +from synapse.replication.slave.storage._slaved_id_tracker import SlavedIdTracker +from synapse.storage.account_data import AccountDataWorkerStore +from synapse.storage.tags import TagsWorkerStore -class SlavedAccountDataStore(BaseSlavedStore): +class SlavedAccountDataStore(TagsWorkerStore, AccountDataWorkerStore, BaseSlavedStore): def __init__(self, db_conn, hs): - super(SlavedAccountDataStore, self).__init__(db_conn, hs) self._account_data_id_gen = SlavedIdTracker( db_conn, "account_data_max_stream_id", "stream_id", ) - self._account_data_stream_cache = StreamChangeCache( - "AccountDataAndTagsChangeCache", - self._account_data_id_gen.get_current_token(), - ) - - get_account_data_for_user = ( - AccountDataStore.__dict__["get_account_data_for_user"] - ) - - get_global_account_data_by_type_for_users = ( - AccountDataStore.__dict__["get_global_account_data_by_type_for_users"] - ) - get_global_account_data_by_type_for_user = ( - AccountDataStore.__dict__["get_global_account_data_by_type_for_user"] - ) - - get_tags_for_user = TagsStore.__dict__["get_tags_for_user"] - get_tags_for_room = ( - DataStore.get_tags_for_room.__func__ - ) - get_account_data_for_room = ( - DataStore.get_account_data_for_room.__func__ - ) - - get_updated_tags = DataStore.get_updated_tags.__func__ - get_updated_account_data_for_user = ( - DataStore.get_updated_account_data_for_user.__func__ - ) + super(SlavedAccountDataStore, self).__init__(db_conn, hs) def get_max_account_data_stream_id(self): return self._account_data_id_gen.get_current_token() @@ -85,6 +56,10 @@ class SlavedAccountDataStore(BaseSlavedStore): (row.data_type, row.user_id,) ) self.get_account_data_for_user.invalidate((row.user_id,)) + self.get_account_data_for_room.invalidate((row.user_id, row.room_id,)) + self.get_account_data_for_room_and_type.invalidate( + (row.user_id, row.room_id, row.data_type,), + ) self._account_data_stream_cache.entity_has_changed( row.user_id, token ) diff --git a/synapse/replication/slave/storage/appservice.py b/synapse/replication/slave/storage/appservice.py index 0d3f31a50c..8cae3076f4 100644 --- a/synapse/replication/slave/storage/appservice.py +++ b/synapse/replication/slave/storage/appservice.py @@ -1,5 +1,6 @@ # -*- coding: utf-8 -*- # Copyright 2015, 2016 OpenMarket Ltd +# Copyright 2018 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. @@ -13,33 +14,11 @@ # See the License for the specific language governing permissions and # limitations under the License. -from ._base import BaseSlavedStore -from synapse.storage import DataStore -from synapse.config.appservice import load_appservices -from synapse.storage.appservice import _make_exclusive_regex +from synapse.storage.appservice import ( + ApplicationServiceWorkerStore, ApplicationServiceTransactionWorkerStore, +) -class SlavedApplicationServiceStore(BaseSlavedStore): - def __init__(self, db_conn, hs): - super(SlavedApplicationServiceStore, self).__init__(db_conn, hs) - self.services_cache = load_appservices( - hs.config.server_name, - hs.config.app_service_config_files - ) - self.exclusive_user_regex = _make_exclusive_regex(self.services_cache) - - get_app_service_by_token = DataStore.get_app_service_by_token.__func__ - get_app_service_by_user_id = DataStore.get_app_service_by_user_id.__func__ - get_app_services = DataStore.get_app_services.__func__ - get_new_events_for_appservice = DataStore.get_new_events_for_appservice.__func__ - create_appservice_txn = DataStore.create_appservice_txn.__func__ - get_appservices_by_state = DataStore.get_appservices_by_state.__func__ - get_oldest_unsent_txn = DataStore.get_oldest_unsent_txn.__func__ - _get_last_txn = DataStore._get_last_txn.__func__ - complete_appservice_txn = DataStore.complete_appservice_txn.__func__ - get_appservice_state = DataStore.get_appservice_state.__func__ - set_appservice_last_pos = DataStore.set_appservice_last_pos.__func__ - set_appservice_state = DataStore.set_appservice_state.__func__ - get_if_app_services_interested_in_user = ( - DataStore.get_if_app_services_interested_in_user.__func__ - ) +class SlavedApplicationServiceStore(ApplicationServiceTransactionWorkerStore, + ApplicationServiceWorkerStore): + pass diff --git a/synapse/replication/slave/storage/directory.py b/synapse/replication/slave/storage/directory.py index 7301d885f2..6deecd3963 100644 --- a/synapse/replication/slave/storage/directory.py +++ b/synapse/replication/slave/storage/directory.py @@ -14,10 +14,8 @@ # limitations under the License. from ._base import BaseSlavedStore -from synapse.storage.directory import DirectoryStore +from synapse.storage.directory import DirectoryWorkerStore -class DirectoryStore(BaseSlavedStore): - get_aliases_for_room = DirectoryStore.__dict__[ - "get_aliases_for_room" - ] +class DirectoryStore(DirectoryWorkerStore, BaseSlavedStore): + pass diff --git a/synapse/replication/slave/storage/events.py b/synapse/replication/slave/storage/events.py index 29d7296b43..b1f64ef0d8 100644 --- a/synapse/replication/slave/storage/events.py +++ b/synapse/replication/slave/storage/events.py @@ -1,5 +1,6 @@ # -*- coding: utf-8 -*- # Copyright 2016 OpenMarket Ltd +# Copyright 2018 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. @@ -15,13 +16,13 @@ import logging from synapse.api.constants import EventTypes -from synapse.storage import DataStore -from synapse.storage.event_federation import EventFederationStore -from synapse.storage.event_push_actions import EventPushActionsStore -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 +from synapse.storage.event_federation import EventFederationWorkerStore +from synapse.storage.event_push_actions import EventPushActionsWorkerStore +from synapse.storage.events_worker import EventsWorkerStore +from synapse.storage.roommember import RoomMemberWorkerStore +from synapse.storage.state import StateGroupWorkerStore +from synapse.storage.stream import StreamWorkerStore +from synapse.storage.signatures import SignatureWorkerStore from ._base import BaseSlavedStore from ._slaved_id_tracker import SlavedIdTracker @@ -37,138 +38,33 @@ logger = logging.getLogger(__name__) # the method descriptor on the DataStore and chuck them into our class. -class SlavedEventStore(StateGroupReadStore, BaseSlavedStore): +class SlavedEventStore(EventFederationWorkerStore, + RoomMemberWorkerStore, + EventPushActionsWorkerStore, + StreamWorkerStore, + EventsWorkerStore, + StateGroupWorkerStore, + SignatureWorkerStore, + BaseSlavedStore): def __init__(self, db_conn, hs): - super(SlavedEventStore, self).__init__(db_conn, hs) self._stream_id_gen = SlavedIdTracker( db_conn, "events", "stream_ordering", ) self._backfill_id_gen = SlavedIdTracker( db_conn, "events", "stream_ordering", step=-1 ) - events_max = self._stream_id_gen.get_current_token() - event_cache_prefill, min_event_val = self._get_cache_dict( - db_conn, "events", - entity_column="room_id", - stream_column="stream_ordering", - max_value=events_max, - ) - self._events_stream_cache = StreamChangeCache( - "EventsRoomStreamChangeCache", min_event_val, - prefilled_cache=event_cache_prefill, - ) - self._membership_stream_cache = StreamChangeCache( - "MembershipStreamChangeCache", events_max, - ) - self.stream_ordering_month_ago = 0 - self._stream_order_on_start = self.get_room_max_stream_ordering() + super(SlavedEventStore, self).__init__(db_conn, hs) # Cached functions can't be accessed through a class instance so we need # to reach inside the __dict__ to extract them. - get_rooms_for_user = RoomMemberStore.__dict__["get_rooms_for_user"] - get_users_in_room = RoomMemberStore.__dict__["get_users_in_room"] - get_hosts_in_room = RoomMemberStore.__dict__["get_hosts_in_room"] - get_users_who_share_room_with_user = ( - RoomMemberStore.__dict__["get_users_who_share_room_with_user"] - ) - get_latest_event_ids_in_room = EventFederationStore.__dict__[ - "get_latest_event_ids_in_room" - ] - get_invited_rooms_for_user = RoomMemberStore.__dict__[ - "get_invited_rooms_for_user" - ] - get_unread_event_push_actions_by_room_for_user = ( - EventPushActionsStore.__dict__["get_unread_event_push_actions_by_room_for_user"] - ) - _get_unread_counts_by_receipt_txn = ( - DataStore._get_unread_counts_by_receipt_txn.__func__ - ) - _get_unread_counts_by_pos_txn = ( - DataStore._get_unread_counts_by_pos_txn.__func__ - ) - get_recent_event_ids_for_room = ( - StreamStore.__dict__["get_recent_event_ids_for_room"] - ) - _get_joined_hosts_cache = RoomMemberStore.__dict__["_get_joined_hosts_cache"] - has_room_changed_since = DataStore.has_room_changed_since.__func__ - - get_unread_push_actions_for_user_in_range_for_http = ( - DataStore.get_unread_push_actions_for_user_in_range_for_http.__func__ - ) - get_unread_push_actions_for_user_in_range_for_email = ( - DataStore.get_unread_push_actions_for_user_in_range_for_email.__func__ - ) - get_push_action_users_in_range = ( - DataStore.get_push_action_users_in_range.__func__ - ) - get_event = DataStore.get_event.__func__ - get_events = DataStore.get_events.__func__ - get_rooms_for_user_where_membership_is = ( - DataStore.get_rooms_for_user_where_membership_is.__func__ - ) - get_membership_changes_for_user = ( - DataStore.get_membership_changes_for_user.__func__ - ) - get_room_events_max_id = DataStore.get_room_events_max_id.__func__ - get_room_events_stream_for_room = ( - DataStore.get_room_events_stream_for_room.__func__ - ) - get_events_around = DataStore.get_events_around.__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 = ( - RoomMemberStore.__dict__["_get_joined_users_from_context"] - ) - - get_joined_hosts = DataStore.get_joined_hosts.__func__ - _get_joined_hosts = RoomMemberStore.__dict__["_get_joined_hosts"] - - get_recent_events_for_room = DataStore.get_recent_events_for_room.__func__ - get_room_events_stream_for_rooms = ( - DataStore.get_room_events_stream_for_rooms.__func__ - ) - is_host_joined = RoomMemberStore.__dict__["is_host_joined"] - get_stream_token_for_event = DataStore.get_stream_token_for_event.__func__ - - _set_before_and_after = staticmethod(DataStore._set_before_and_after) - - _get_events = DataStore._get_events.__func__ - _get_events_from_cache = DataStore._get_events_from_cache.__func__ - - _invalidate_get_event_cache = DataStore._invalidate_get_event_cache.__func__ - _enqueue_events = DataStore._enqueue_events.__func__ - _do_fetch = DataStore._do_fetch.__func__ - _fetch_event_rows = DataStore._fetch_event_rows.__func__ - _get_event_from_row = DataStore._get_event_from_row.__func__ - _get_rooms_for_user_where_membership_is_txn = ( - DataStore._get_rooms_for_user_where_membership_is_txn.__func__ - ) - _get_events_around_txn = DataStore._get_events_around_txn.__func__ - - get_backfill_events = DataStore.get_backfill_events.__func__ - _get_backfill_events = DataStore._get_backfill_events.__func__ - get_missing_events = DataStore.get_missing_events.__func__ - _get_missing_events = DataStore._get_missing_events.__func__ - - get_auth_chain = DataStore.get_auth_chain.__func__ - get_auth_chain_ids = DataStore.get_auth_chain_ids.__func__ - _get_auth_chain_ids_txn = DataStore._get_auth_chain_ids_txn.__func__ - - get_room_max_stream_ordering = DataStore.get_room_max_stream_ordering.__func__ - - get_forward_extremeties_for_room = ( - DataStore.get_forward_extremeties_for_room.__func__ - ) - _get_forward_extremeties_for_room = ( - EventFederationStore.__dict__["_get_forward_extremeties_for_room"] - ) - - get_all_new_events_stream = DataStore.get_all_new_events_stream.__func__ - - get_federation_out_pos = DataStore.get_federation_out_pos.__func__ - update_federation_out_pos = DataStore.update_federation_out_pos.__func__ + + def get_room_max_stream_ordering(self): + return self._stream_id_gen.get_current_token() + + def get_room_min_stream_ordering(self): + return self._backfill_id_gen.get_current_token() def stream_positions(self): result = super(SlavedEventStore, self).stream_positions() diff --git a/synapse/replication/slave/storage/profile.py b/synapse/replication/slave/storage/profile.py new file mode 100644 index 0000000000..46c28d4171 --- /dev/null +++ b/synapse/replication/slave/storage/profile.py @@ -0,0 +1,21 @@ +# -*- coding: utf-8 -*- +# Copyright 2018 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. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +from synapse.replication.slave.storage._base import BaseSlavedStore +from synapse.storage.profile import ProfileWorkerStore + + +class SlavedProfileStore(ProfileWorkerStore, BaseSlavedStore): + pass diff --git a/synapse/replication/slave/storage/push_rule.py b/synapse/replication/slave/storage/push_rule.py index 83e880fdd2..bb2c40b6e3 100644 --- a/synapse/replication/slave/storage/push_rule.py +++ b/synapse/replication/slave/storage/push_rule.py @@ -1,5 +1,6 @@ # -*- coding: utf-8 -*- # Copyright 2015, 2016 OpenMarket Ltd +# Copyright 2018 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. @@ -15,29 +16,15 @@ from .events import SlavedEventStore from ._slaved_id_tracker import SlavedIdTracker -from synapse.storage import DataStore -from synapse.storage.push_rule import PushRuleStore -from synapse.util.caches.stream_change_cache import StreamChangeCache +from synapse.storage.push_rule import PushRulesWorkerStore -class SlavedPushRuleStore(SlavedEventStore): +class SlavedPushRuleStore(PushRulesWorkerStore, SlavedEventStore): def __init__(self, db_conn, hs): - super(SlavedPushRuleStore, self).__init__(db_conn, hs) self._push_rules_stream_id_gen = SlavedIdTracker( db_conn, "push_rules_stream", "stream_id", ) - self.push_rules_stream_cache = StreamChangeCache( - "PushRulesStreamChangeCache", - self._push_rules_stream_id_gen.get_current_token(), - ) - - get_push_rules_for_user = PushRuleStore.__dict__["get_push_rules_for_user"] - get_push_rules_enabled_for_user = ( - PushRuleStore.__dict__["get_push_rules_enabled_for_user"] - ) - have_push_rules_changed_for_user = ( - DataStore.have_push_rules_changed_for_user.__func__ - ) + super(SlavedPushRuleStore, self).__init__(db_conn, hs) def get_push_rules_stream_token(self): return ( @@ -45,6 +32,9 @@ class SlavedPushRuleStore(SlavedEventStore): self._stream_id_gen.get_current_token(), ) + def get_max_push_rules_stream_id(self): + return self._push_rules_stream_id_gen.get_current_token() + def stream_positions(self): result = super(SlavedPushRuleStore, self).stream_positions() result["push_rules"] = self._push_rules_stream_id_gen.get_current_token() diff --git a/synapse/replication/slave/storage/pushers.py b/synapse/replication/slave/storage/pushers.py index 4e8d68ece9..a7cd5a7291 100644 --- a/synapse/replication/slave/storage/pushers.py +++ b/synapse/replication/slave/storage/pushers.py @@ -1,5 +1,6 @@ # -*- coding: utf-8 -*- # Copyright 2016 OpenMarket Ltd +# Copyright 2018 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. @@ -16,10 +17,10 @@ from ._base import BaseSlavedStore from ._slaved_id_tracker import SlavedIdTracker -from synapse.storage import DataStore +from synapse.storage.pusher import PusherWorkerStore -class SlavedPusherStore(BaseSlavedStore): +class SlavedPusherStore(PusherWorkerStore, BaseSlavedStore): def __init__(self, db_conn, hs): super(SlavedPusherStore, self).__init__(db_conn, hs) @@ -28,13 +29,6 @@ class SlavedPusherStore(BaseSlavedStore): extra_tables=[("deleted_pushers", "stream_id")], ) - get_all_pushers = DataStore.get_all_pushers.__func__ - get_pushers_by = DataStore.get_pushers_by.__func__ - get_pushers_by_app_id_and_pushkey = ( - DataStore.get_pushers_by_app_id_and_pushkey.__func__ - ) - _decode_pushers_rows = DataStore._decode_pushers_rows.__func__ - def stream_positions(self): result = super(SlavedPusherStore, self).stream_positions() result["pushers"] = self._pushers_id_gen.get_current_token() diff --git a/synapse/replication/slave/storage/receipts.py b/synapse/replication/slave/storage/receipts.py index b371574ece..1647072f65 100644 --- a/synapse/replication/slave/storage/receipts.py +++ b/synapse/replication/slave/storage/receipts.py @@ -1,5 +1,6 @@ # -*- coding: utf-8 -*- # Copyright 2016 OpenMarket Ltd +# Copyright 2018 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. @@ -16,9 +17,7 @@ from ._base import BaseSlavedStore from ._slaved_id_tracker import SlavedIdTracker -from synapse.storage import DataStore -from synapse.storage.receipts import ReceiptsStore -from synapse.util.caches.stream_change_cache import StreamChangeCache +from synapse.storage.receipts import ReceiptsWorkerStore # So, um, we want to borrow a load of functions intended for reading from # a DataStore, but we don't want to take functions that either write to the @@ -29,36 +28,19 @@ from synapse.util.caches.stream_change_cache import StreamChangeCache # the method descriptor on the DataStore and chuck them into our class. -class SlavedReceiptsStore(BaseSlavedStore): +class SlavedReceiptsStore(ReceiptsWorkerStore, BaseSlavedStore): def __init__(self, db_conn, hs): - super(SlavedReceiptsStore, self).__init__(db_conn, hs) - + # We instantiate this first as the ReceiptsWorkerStore constructor + # needs to be able to call get_max_receipt_stream_id self._receipts_id_gen = SlavedIdTracker( db_conn, "receipts_linearized", "stream_id" ) - self._receipts_stream_cache = StreamChangeCache( - "ReceiptsRoomChangeCache", self._receipts_id_gen.get_current_token() - ) - - get_receipts_for_user = ReceiptsStore.__dict__["get_receipts_for_user"] - get_linearized_receipts_for_room = ( - ReceiptsStore.__dict__["get_linearized_receipts_for_room"] - ) - _get_linearized_receipts_for_rooms = ( - ReceiptsStore.__dict__["_get_linearized_receipts_for_rooms"] - ) - get_last_receipt_event_id_for_user = ( - ReceiptsStore.__dict__["get_last_receipt_event_id_for_user"] - ) - - get_max_receipt_stream_id = DataStore.get_max_receipt_stream_id.__func__ - get_all_updated_receipts = DataStore.get_all_updated_receipts.__func__ + super(SlavedReceiptsStore, self).__init__(db_conn, hs) - get_linearized_receipts_for_rooms = ( - DataStore.get_linearized_receipts_for_rooms.__func__ - ) + def get_max_receipt_stream_id(self): + return self._receipts_id_gen.get_current_token() def stream_positions(self): result = super(SlavedReceiptsStore, self).stream_positions() @@ -71,6 +53,8 @@ class SlavedReceiptsStore(BaseSlavedStore): self.get_last_receipt_event_id_for_user.invalidate( (user_id, room_id, receipt_type) ) + self._invalidate_get_users_with_receipts_in_room(room_id, receipt_type, user_id) + self.get_receipts_for_room.invalidate((room_id, receipt_type)) def process_replication_rows(self, stream_name, token, rows): if stream_name == "receipts": diff --git a/synapse/replication/slave/storage/registration.py b/synapse/replication/slave/storage/registration.py index e27c7332d2..7323bf0f1e 100644 --- a/synapse/replication/slave/storage/registration.py +++ b/synapse/replication/slave/storage/registration.py @@ -14,20 +14,8 @@ # limitations under the License. from ._base import BaseSlavedStore -from synapse.storage import DataStore -from synapse.storage.registration import RegistrationStore +from synapse.storage.registration import RegistrationWorkerStore -class SlavedRegistrationStore(BaseSlavedStore): - def __init__(self, db_conn, hs): - super(SlavedRegistrationStore, self).__init__(db_conn, hs) - - # TODO: use the cached version and invalidate deleted tokens - get_user_by_access_token = RegistrationStore.__dict__[ - "get_user_by_access_token" - ] - - _query_for_auth = DataStore._query_for_auth.__func__ - get_user_by_id = RegistrationStore.__dict__[ - "get_user_by_id" - ] +class SlavedRegistrationStore(RegistrationWorkerStore, BaseSlavedStore): + pass diff --git a/synapse/replication/slave/storage/room.py b/synapse/replication/slave/storage/room.py index f510384033..5ae1670157 100644 --- a/synapse/replication/slave/storage/room.py +++ b/synapse/replication/slave/storage/room.py @@ -14,32 +14,19 @@ # limitations under the License. from ._base import BaseSlavedStore -from synapse.storage import DataStore -from synapse.storage.room import RoomStore +from synapse.storage.room import RoomWorkerStore from ._slaved_id_tracker import SlavedIdTracker -class RoomStore(BaseSlavedStore): +class RoomStore(RoomWorkerStore, BaseSlavedStore): def __init__(self, db_conn, hs): super(RoomStore, self).__init__(db_conn, hs) self._public_room_id_gen = SlavedIdTracker( db_conn, "public_room_list_stream", "stream_id" ) - get_public_room_ids = DataStore.get_public_room_ids.__func__ - get_current_public_room_stream_id = ( - DataStore.get_current_public_room_stream_id.__func__ - ) - get_public_room_ids_at_stream_id = ( - RoomStore.__dict__["get_public_room_ids_at_stream_id"] - ) - get_public_room_ids_at_stream_id_txn = ( - DataStore.get_public_room_ids_at_stream_id_txn.__func__ - ) - get_published_at_stream_id_txn = ( - DataStore.get_published_at_stream_id_txn.__func__ - ) - get_public_room_changes = DataStore.get_public_room_changes.__func__ + def get_current_public_room_stream_id(self): + return self._public_room_id_gen.get_current_token() def stream_positions(self): result = super(RoomStore, self).stream_positions() diff --git a/synapse/replication/tcp/commands.py b/synapse/replication/tcp/commands.py index 9633404f73..0005ad5879 100644 --- a/synapse/replication/tcp/commands.py +++ b/synapse/replication/tcp/commands.py @@ -19,7 +19,7 @@ allowed to be sent by which side. """ import logging -import simplejson as json +import simplejson logger = logging.getLogger(__name__) @@ -100,14 +100,14 @@ class RdataCommand(Command): return cls( stream_name, None if token == "batch" else int(token), - json.loads(row_json) + simplejson.loads(row_json) ) def to_line(self): return " ".join(( self.stream_name, str(self.token) if self.token is not None else "batch", - json.dumps(self.row), + simplejson.dumps(self.row, namedtuple_as_object=False), )) @@ -298,10 +298,12 @@ class InvalidateCacheCommand(Command): def from_line(cls, line): cache_func, keys_json = line.split(" ", 1) - return cls(cache_func, json.loads(keys_json)) + return cls(cache_func, simplejson.loads(keys_json)) def to_line(self): - return " ".join((self.cache_func, json.dumps(self.keys))) + return " ".join(( + self.cache_func, simplejson.dumps(self.keys, namedtuple_as_object=False) + )) class UserIpCommand(Command): @@ -325,14 +327,14 @@ class UserIpCommand(Command): def from_line(cls, line): user_id, jsn = line.split(" ", 1) - access_token, ip, user_agent, device_id, last_seen = json.loads(jsn) + access_token, ip, user_agent, device_id, last_seen = simplejson.loads(jsn) return cls( user_id, access_token, ip, user_agent, device_id, last_seen ) def to_line(self): - return self.user_id + " " + json.dumps(( + return self.user_id + " " + simplejson.dumps(( self.access_token, self.ip, self.user_agent, self.device_id, self.last_seen, )) diff --git a/synapse/replication/tcp/protocol.py b/synapse/replication/tcp/protocol.py index d59503b905..0a9a290af4 100644 --- a/synapse/replication/tcp/protocol.py +++ b/synapse/replication/tcp/protocol.py @@ -517,25 +517,28 @@ class ClientReplicationStreamProtocol(BaseReplicationStreamProtocol): self.send_error("Wrong remote") def on_RDATA(self, cmd): + stream_name = cmd.stream_name + inbound_rdata_count.inc(stream_name) + try: - row = STREAMS_MAP[cmd.stream_name].ROW_TYPE(*cmd.row) + row = STREAMS_MAP[stream_name].ROW_TYPE(*cmd.row) except Exception: logger.exception( "[%s] Failed to parse RDATA: %r %r", - self.id(), cmd.stream_name, cmd.row + self.id(), stream_name, cmd.row ) raise if cmd.token is None: # I.e. this is part of a batch of updates for this stream. Batch # until we get an update for the stream with a non None token - self.pending_batches.setdefault(cmd.stream_name, []).append(row) + self.pending_batches.setdefault(stream_name, []).append(row) else: # Check if this is the last of a batch of updates - rows = self.pending_batches.pop(cmd.stream_name, []) + rows = self.pending_batches.pop(stream_name, []) rows.append(row) - self.handler.on_rdata(cmd.stream_name, cmd.token, rows) + self.handler.on_rdata(stream_name, cmd.token, rows) def on_POSITION(self, cmd): self.handler.on_position(cmd.stream_name, cmd.token) @@ -644,3 +647,9 @@ metrics.register_callback( }, labels=["command", "name", "conn_id"], ) + +# number of updates received for each RDATA stream +inbound_rdata_count = metrics.register_counter( + "inbound_rdata_count", + labels=["stream_name"], +) diff --git a/synapse/rest/client/v1/admin.py b/synapse/rest/client/v1/admin.py index 5022808ea9..303419d281 100644 --- a/synapse/rest/client/v1/admin.py +++ b/synapse/rest/client/v1/admin.py @@ -1,5 +1,6 @@ # -*- coding: utf-8 -*- # Copyright 2014-2016 OpenMarket Ltd +# Copyright 2018 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. @@ -16,7 +17,7 @@ from twisted.internet import defer from synapse.api.constants import Membership -from synapse.api.errors import AuthError, SynapseError +from synapse.api.errors import AuthError, SynapseError, Codes, NotFoundError from synapse.types import UserID, create_requester from synapse.http.servlet import parse_json_object_from_request @@ -113,12 +114,18 @@ class PurgeMediaCacheRestServlet(ClientV1RestServlet): class PurgeHistoryRestServlet(ClientV1RestServlet): PATTERNS = client_path_patterns( - "/admin/purge_history/(?P<room_id>[^/]*)/(?P<event_id>[^/]*)" + "/admin/purge_history/(?P<room_id>[^/]*)(/(?P<event_id>[^/]+))?" ) def __init__(self, hs): + """ + + Args: + hs (synapse.server.HomeServer) + """ super(PurgeHistoryRestServlet, self).__init__(hs) self.handlers = hs.get_handlers() + self.store = hs.get_datastore() @defer.inlineCallbacks def on_POST(self, request, room_id, event_id): @@ -128,9 +135,93 @@ class PurgeHistoryRestServlet(ClientV1RestServlet): if not is_admin: raise AuthError(403, "You are not a server admin") - yield self.handlers.message_handler.purge_history(room_id, event_id) + body = parse_json_object_from_request(request, allow_empty_body=True) - defer.returnValue((200, {})) + delete_local_events = bool(body.get("delete_local_events", False)) + + # establish the topological ordering we should keep events from. The + # user can provide an event_id in the URL or the request body, or can + # provide a timestamp in the request body. + if event_id is None: + event_id = body.get('purge_up_to_event_id') + + if event_id is not None: + event = yield self.store.get_event(event_id) + + if event.room_id != room_id: + raise SynapseError(400, "Event is for wrong room.") + + depth = event.depth + logger.info( + "[purge] purging up to depth %i (event_id %s)", + depth, event_id, + ) + elif 'purge_up_to_ts' in body: + ts = body['purge_up_to_ts'] + if not isinstance(ts, int): + raise SynapseError( + 400, "purge_up_to_ts must be an int", + errcode=Codes.BAD_JSON, + ) + + stream_ordering = ( + yield self.store.find_first_stream_ordering_after_ts(ts) + ) + + (_, depth, _) = ( + yield self.store.get_room_event_after_stream_ordering( + room_id, stream_ordering, + ) + ) + logger.info( + "[purge] purging up to depth %i (received_ts %i => " + "stream_ordering %i)", + depth, ts, stream_ordering, + ) + else: + raise SynapseError( + 400, + "must specify purge_up_to_event_id or purge_up_to_ts", + errcode=Codes.BAD_JSON, + ) + + purge_id = yield self.handlers.message_handler.start_purge_history( + room_id, depth, + delete_local_events=delete_local_events, + ) + + defer.returnValue((200, { + "purge_id": purge_id, + })) + + +class PurgeHistoryStatusRestServlet(ClientV1RestServlet): + PATTERNS = client_path_patterns( + "/admin/purge_history_status/(?P<purge_id>[^/]+)" + ) + + def __init__(self, hs): + """ + + Args: + hs (synapse.server.HomeServer) + """ + super(PurgeHistoryStatusRestServlet, self).__init__(hs) + self.handlers = hs.get_handlers() + + @defer.inlineCallbacks + def on_GET(self, request, purge_id): + requester = yield self.auth.get_user_by_req(request) + is_admin = yield self.auth.is_server_admin(requester.user) + + if not is_admin: + raise AuthError(403, "You are not a server admin") + + purge_status = self.handlers.message_handler.get_purge_status(purge_id) + if purge_status is None: + raise NotFoundError("purge id '%s' not found" % purge_id) + + defer.returnValue((200, purge_status.asdict())) class DeactivateAccountRestServlet(ClientV1RestServlet): @@ -171,6 +262,8 @@ class ShutdownRoomRestServlet(ClientV1RestServlet): self.store = hs.get_datastore() self.handlers = hs.get_handlers() self.state = hs.get_state_handler() + self.event_creation_handler = hs.get_event_creation_handler() + self.room_member_handler = hs.get_room_member_handler() @defer.inlineCallbacks def on_POST(self, request, room_id): @@ -203,8 +296,7 @@ class ShutdownRoomRestServlet(ClientV1RestServlet): ) new_room_id = info["room_id"] - msg_handler = self.handlers.message_handler - yield msg_handler.create_and_send_nonmember_event( + yield self.event_creation_handler.create_and_send_nonmember_event( room_creator_requester, { "type": "m.room.message", @@ -230,7 +322,7 @@ class ShutdownRoomRestServlet(ClientV1RestServlet): logger.info("Kicking %r from %r...", user_id, room_id) target_requester = create_requester(user_id) - yield self.handlers.room_member_handler.update_membership( + yield self.room_member_handler.update_membership( requester=target_requester, target=target_requester.user, room_id=room_id, @@ -239,9 +331,9 @@ class ShutdownRoomRestServlet(ClientV1RestServlet): ratelimit=False ) - yield self.handlers.room_member_handler.forget(target_requester.user, room_id) + yield self.room_member_handler.forget(target_requester.user, room_id) - yield self.handlers.room_member_handler.update_membership( + yield self.room_member_handler.update_membership( requester=target_requester, target=target_requester.user, room_id=new_room_id, @@ -289,6 +381,27 @@ class QuarantineMediaInRoom(ClientV1RestServlet): defer.returnValue((200, {"num_quarantined": num_quarantined})) +class ListMediaInRoom(ClientV1RestServlet): + """Lists all of the media in a given room. + """ + PATTERNS = client_path_patterns("/admin/room/(?P<room_id>[^/]+)/media") + + def __init__(self, hs): + super(ListMediaInRoom, self).__init__(hs) + self.store = hs.get_datastore() + + @defer.inlineCallbacks + def on_GET(self, request, room_id): + requester = yield self.auth.get_user_by_req(request) + is_admin = yield self.auth.is_server_admin(requester.user) + if not is_admin: + raise AuthError(403, "You are not a server admin") + + local_mxcs, remote_mxcs = yield self.store.get_media_mxcs_in_room(room_id) + + defer.returnValue((200, {"local": local_mxcs, "remote": remote_mxcs})) + + class ResetPasswordRestServlet(ClientV1RestServlet): """Post request to allow an administrator reset password for a user. This needs user to have administrator access in Synapse. @@ -479,6 +592,7 @@ class SearchUsersRestServlet(ClientV1RestServlet): def register_servlets(hs, http_server): WhoisRestServlet(hs).register(http_server) PurgeMediaCacheRestServlet(hs).register(http_server) + PurgeHistoryStatusRestServlet(hs).register(http_server) DeactivateAccountRestServlet(hs).register(http_server) PurgeHistoryRestServlet(hs).register(http_server) UsersRestServlet(hs).register(http_server) @@ -487,3 +601,4 @@ def register_servlets(hs, http_server): SearchUsersRestServlet(hs).register(http_server) ShutdownRoomRestServlet(hs).register(http_server) QuarantineMediaInRoom(hs).register(http_server) + ListMediaInRoom(hs).register(http_server) diff --git a/synapse/rest/client/v1/login.py b/synapse/rest/client/v1/login.py index 5669ecb724..45844aa2d2 100644 --- a/synapse/rest/client/v1/login.py +++ b/synapse/rest/client/v1/login.py @@ -191,19 +191,25 @@ class LoginRestServlet(ClientV1RestServlet): # convert threepid identifiers to user IDs if identifier["type"] == "m.id.thirdparty": - if 'medium' not in identifier or 'address' not in identifier: + address = identifier.get('address') + medium = identifier.get('medium') + + if medium is None or address is None: raise SynapseError(400, "Invalid thirdparty identifier") - address = identifier['address'] - if identifier['medium'] == 'email': + if medium == 'email': # For emails, transform the address to lowercase. # We store all email addreses as lowercase in the DB. # (See add_threepid in synapse/handlers/auth.py) address = address.lower() user_id = yield self.hs.get_datastore().get_user_id_by_threepid( - identifier['medium'], address + medium, address, ) if not user_id: + logger.warn( + "unknown 3pid identifier medium %s, address %r", + medium, address, + ) raise LoginError(403, "", errcode=Codes.FORBIDDEN) identifier = { diff --git a/synapse/rest/client/v1/register.py b/synapse/rest/client/v1/register.py index 32ed1d3ab2..5c5fa8f7ab 100644 --- a/synapse/rest/client/v1/register.py +++ b/synapse/rest/client/v1/register.py @@ -70,10 +70,15 @@ class RegisterRestServlet(ClientV1RestServlet): self.handlers = hs.get_handlers() def on_GET(self, request): + + require_email = 'email' in self.hs.config.registrations_require_3pid + require_msisdn = 'msisdn' in self.hs.config.registrations_require_3pid + + flows = [] if self.hs.config.enable_registration_captcha: - return ( - 200, - {"flows": [ + # only support the email-only flow if we don't require MSISDN 3PIDs + if not require_msisdn: + flows.extend([ { "type": LoginType.RECAPTCHA, "stages": [ @@ -82,27 +87,34 @@ class RegisterRestServlet(ClientV1RestServlet): LoginType.PASSWORD ] }, + ]) + # only support 3PIDless registration if no 3PIDs are required + if not require_email and not require_msisdn: + flows.extend([ { "type": LoginType.RECAPTCHA, "stages": [LoginType.RECAPTCHA, LoginType.PASSWORD] } - ]} - ) + ]) else: - return ( - 200, - {"flows": [ + # only support the email-only flow if we don't require MSISDN 3PIDs + if require_email or not require_msisdn: + flows.extend([ { "type": LoginType.EMAIL_IDENTITY, "stages": [ LoginType.EMAIL_IDENTITY, LoginType.PASSWORD ] - }, + } + ]) + # only support 3PIDless registration if no 3PIDs are required + if not require_email and not require_msisdn: + flows.extend([ { "type": LoginType.PASSWORD } - ]} - ) + ]) + return (200, {"flows": flows}) @defer.inlineCallbacks def on_POST(self, request): diff --git a/synapse/rest/client/v1/room.py b/synapse/rest/client/v1/room.py index 80989731fa..70d788deea 100644 --- a/synapse/rest/client/v1/room.py +++ b/synapse/rest/client/v1/room.py @@ -1,5 +1,6 @@ # -*- coding: utf-8 -*- # Copyright 2014-2016 OpenMarket Ltd +# Copyright 2018 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. @@ -82,6 +83,8 @@ class RoomStateEventRestServlet(ClientV1RestServlet): def __init__(self, hs): super(RoomStateEventRestServlet, self).__init__(hs) self.handlers = hs.get_handlers() + self.event_creation_hander = hs.get_event_creation_handler() + self.room_member_handler = hs.get_room_member_handler() def register(self, http_server): # /room/$roomid/state/$eventtype @@ -154,7 +157,7 @@ class RoomStateEventRestServlet(ClientV1RestServlet): if event_type == EventTypes.Member: membership = content.get("membership", None) - event = yield self.handlers.room_member_handler.update_membership( + event = yield self.room_member_handler.update_membership( requester, target=UserID.from_string(state_key), room_id=room_id, @@ -162,15 +165,16 @@ class RoomStateEventRestServlet(ClientV1RestServlet): content=content, ) else: - msg_handler = self.handlers.message_handler - event, context = yield msg_handler.create_event( + event, context = yield self.event_creation_hander.create_event( requester, event_dict, token_id=requester.access_token_id, txn_id=txn_id, ) - yield msg_handler.send_nonmember_event(requester, event, context) + yield self.event_creation_hander.send_nonmember_event( + requester, event, context, + ) ret = {} if event: @@ -183,7 +187,7 @@ class RoomSendEventRestServlet(ClientV1RestServlet): def __init__(self, hs): super(RoomSendEventRestServlet, self).__init__(hs) - self.handlers = hs.get_handlers() + self.event_creation_hander = hs.get_event_creation_handler() def register(self, http_server): # /rooms/$roomid/send/$event_type[/$txn_id] @@ -195,15 +199,19 @@ class RoomSendEventRestServlet(ClientV1RestServlet): requester = yield self.auth.get_user_by_req(request, allow_guest=True) content = parse_json_object_from_request(request) - msg_handler = self.handlers.message_handler - event = yield msg_handler.create_and_send_nonmember_event( + event_dict = { + "type": event_type, + "content": content, + "room_id": room_id, + "sender": requester.user.to_string(), + } + + if 'ts' in request.args and requester.app_service: + event_dict['origin_server_ts'] = parse_integer(request, "ts", 0) + + event = yield self.event_creation_hander.create_and_send_nonmember_event( requester, - { - "type": event_type, - "content": content, - "room_id": room_id, - "sender": requester.user.to_string(), - }, + event_dict, txn_id=txn_id, ) @@ -222,7 +230,7 @@ class RoomSendEventRestServlet(ClientV1RestServlet): class JoinRoomAliasServlet(ClientV1RestServlet): def __init__(self, hs): super(JoinRoomAliasServlet, self).__init__(hs) - self.handlers = hs.get_handlers() + self.room_member_handler = hs.get_room_member_handler() def register(self, http_server): # /join/$room_identifier[/$txn_id] @@ -250,7 +258,7 @@ class JoinRoomAliasServlet(ClientV1RestServlet): except Exception: remote_room_hosts = None elif RoomAlias.is_valid(room_identifier): - handler = self.handlers.room_member_handler + handler = self.room_member_handler room_alias = RoomAlias.from_string(room_identifier) room_id, remote_room_hosts = yield handler.lookup_room_alias(room_alias) room_id = room_id.to_string() @@ -259,7 +267,7 @@ class JoinRoomAliasServlet(ClientV1RestServlet): room_identifier, )) - yield self.handlers.room_member_handler.update_membership( + yield self.room_member_handler.update_membership( requester=requester, target=requester.user, room_id=room_id, @@ -487,13 +495,35 @@ class RoomInitialSyncRestServlet(ClientV1RestServlet): defer.returnValue((200, content)) -class RoomEventContext(ClientV1RestServlet): +class RoomEventServlet(ClientV1RestServlet): + PATTERNS = client_path_patterns( + "/rooms/(?P<room_id>[^/]*)/event/(?P<event_id>[^/]*)$" + ) + + def __init__(self, hs): + super(RoomEventServlet, self).__init__(hs) + self.clock = hs.get_clock() + self.event_handler = hs.get_event_handler() + + @defer.inlineCallbacks + def on_GET(self, request, room_id, event_id): + requester = yield self.auth.get_user_by_req(request) + event = yield self.event_handler.get_event(requester.user, event_id) + + time_now = self.clock.time_msec() + if event: + defer.returnValue((200, serialize_event(event, time_now))) + else: + defer.returnValue((404, "Event not found.")) + + +class RoomEventContextServlet(ClientV1RestServlet): PATTERNS = client_path_patterns( "/rooms/(?P<room_id>[^/]*)/context/(?P<event_id>[^/]*)$" ) def __init__(self, hs): - super(RoomEventContext, self).__init__(hs) + super(RoomEventContextServlet, self).__init__(hs) self.clock = hs.get_clock() self.handlers = hs.get_handlers() @@ -533,7 +563,7 @@ class RoomEventContext(ClientV1RestServlet): class RoomForgetRestServlet(ClientV1RestServlet): def __init__(self, hs): super(RoomForgetRestServlet, self).__init__(hs) - self.handlers = hs.get_handlers() + self.room_member_handler = hs.get_room_member_handler() def register(self, http_server): PATTERNS = ("/rooms/(?P<room_id>[^/]*)/forget") @@ -546,7 +576,7 @@ class RoomForgetRestServlet(ClientV1RestServlet): allow_guest=False, ) - yield self.handlers.room_member_handler.forget( + yield self.room_member_handler.forget( user=requester.user, room_id=room_id, ) @@ -564,12 +594,12 @@ class RoomMembershipRestServlet(ClientV1RestServlet): def __init__(self, hs): super(RoomMembershipRestServlet, self).__init__(hs) - self.handlers = hs.get_handlers() + self.room_member_handler = hs.get_room_member_handler() def register(self, http_server): # /rooms/$roomid/[invite|join|leave] PATTERNS = ("/rooms/(?P<room_id>[^/]*)/" - "(?P<membership_action>join|invite|leave|ban|unban|kick|forget)") + "(?P<membership_action>join|invite|leave|ban|unban|kick)") register_txn_path(self, PATTERNS, http_server) @defer.inlineCallbacks @@ -593,7 +623,7 @@ class RoomMembershipRestServlet(ClientV1RestServlet): content = {} if membership_action == "invite" and self._has_3pid_invite_keys(content): - yield self.handlers.room_member_handler.do_3pid_invite( + yield self.room_member_handler.do_3pid_invite( room_id, requester.user, content["medium"], @@ -615,7 +645,7 @@ class RoomMembershipRestServlet(ClientV1RestServlet): if 'reason' in content and membership_action in ['kick', 'ban']: event_content = {'reason': content['reason']} - yield self.handlers.room_member_handler.update_membership( + yield self.room_member_handler.update_membership( requester=requester, target=target, room_id=room_id, @@ -643,6 +673,7 @@ class RoomRedactEventRestServlet(ClientV1RestServlet): def __init__(self, hs): super(RoomRedactEventRestServlet, self).__init__(hs) self.handlers = hs.get_handlers() + self.event_creation_handler = hs.get_event_creation_handler() def register(self, http_server): PATTERNS = ("/rooms/(?P<room_id>[^/]*)/redact/(?P<event_id>[^/]*)") @@ -653,8 +684,7 @@ class RoomRedactEventRestServlet(ClientV1RestServlet): requester = yield self.auth.get_user_by_req(request) content = parse_json_object_from_request(request) - msg_handler = self.handlers.message_handler - event = yield msg_handler.create_and_send_nonmember_event( + event = yield self.event_creation_handler.create_and_send_nonmember_event( requester, { "type": EventTypes.Redaction, @@ -803,4 +833,5 @@ def register_servlets(hs, http_server): RoomTypingRestServlet(hs).register(http_server) SearchRestServlet(hs).register(http_server) JoinedRoomsRestServlet(hs).register(http_server) - RoomEventContext(hs).register(http_server) + RoomEventServlet(hs).register(http_server) + RoomEventContextServlet(hs).register(http_server) diff --git a/synapse/rest/client/v2_alpha/account.py b/synapse/rest/client/v2_alpha/account.py index 385a3ad2ec..30523995af 100644 --- a/synapse/rest/client/v2_alpha/account.py +++ b/synapse/rest/client/v2_alpha/account.py @@ -26,6 +26,7 @@ from synapse.http.servlet import ( ) from synapse.util.async import run_on_reactor from synapse.util.msisdn import phone_number_to_msisdn +from synapse.util.threepids import check_3pid_allowed from ._base import client_v2_patterns, interactive_auth_handler logger = logging.getLogger(__name__) @@ -47,6 +48,11 @@ class EmailPasswordRequestTokenRestServlet(RestServlet): 'id_server', 'client_secret', 'email', 'send_attempt' ]) + if not check_3pid_allowed(self.hs, "email", body['email']): + raise SynapseError( + 403, "Third party identifier is not allowed", Codes.THREEPID_DENIED, + ) + existingUid = yield self.hs.get_datastore().get_user_id_by_threepid( 'email', body['email'] ) @@ -78,6 +84,11 @@ class MsisdnPasswordRequestTokenRestServlet(RestServlet): msisdn = phone_number_to_msisdn(body['country'], body['phone_number']) + if not check_3pid_allowed(self.hs, "msisdn", msisdn): + raise SynapseError( + 403, "Third party identifier is not allowed", Codes.THREEPID_DENIED, + ) + existingUid = yield self.datastore.get_user_id_by_threepid( 'msisdn', msisdn ) @@ -217,6 +228,11 @@ class EmailThreepidRequestTokenRestServlet(RestServlet): if absent: raise SynapseError(400, "Missing params: %r" % absent, Codes.MISSING_PARAM) + if not check_3pid_allowed(self.hs, "email", body['email']): + raise SynapseError( + 403, "Third party identifier is not allowed", Codes.THREEPID_DENIED, + ) + existingUid = yield self.datastore.get_user_id_by_threepid( 'email', body['email'] ) @@ -255,6 +271,11 @@ class MsisdnThreepidRequestTokenRestServlet(RestServlet): msisdn = phone_number_to_msisdn(body['country'], body['phone_number']) + if not check_3pid_allowed(self.hs, "msisdn", msisdn): + raise SynapseError( + 403, "Third party identifier is not allowed", Codes.THREEPID_DENIED, + ) + existingUid = yield self.datastore.get_user_id_by_threepid( 'msisdn', msisdn ) diff --git a/synapse/rest/client/v2_alpha/register.py b/synapse/rest/client/v2_alpha/register.py index e9d88a8895..0ba62bddc1 100644 --- a/synapse/rest/client/v2_alpha/register.py +++ b/synapse/rest/client/v2_alpha/register.py @@ -26,6 +26,7 @@ from synapse.http.servlet import ( RestServlet, parse_json_object_from_request, assert_params_in_request, parse_string ) from synapse.util.msisdn import phone_number_to_msisdn +from synapse.util.threepids import check_3pid_allowed from ._base import client_v2_patterns, interactive_auth_handler @@ -70,6 +71,11 @@ class EmailRegisterRequestTokenRestServlet(RestServlet): 'id_server', 'client_secret', 'email', 'send_attempt' ]) + if not check_3pid_allowed(self.hs, "email", body['email']): + raise SynapseError( + 403, "Third party identifier is not allowed", Codes.THREEPID_DENIED, + ) + existingUid = yield self.hs.get_datastore().get_user_id_by_threepid( 'email', body['email'] ) @@ -105,6 +111,11 @@ class MsisdnRegisterRequestTokenRestServlet(RestServlet): msisdn = phone_number_to_msisdn(body['country'], body['phone_number']) + if not check_3pid_allowed(self.hs, "msisdn", msisdn): + raise SynapseError( + 403, "Third party identifier is not allowed", Codes.THREEPID_DENIED, + ) + existingUid = yield self.hs.get_datastore().get_user_id_by_threepid( 'msisdn', msisdn ) @@ -172,7 +183,7 @@ class RegisterRestServlet(RestServlet): self.auth_handler = hs.get_auth_handler() self.registration_handler = hs.get_handlers().registration_handler self.identity_handler = hs.get_handlers().identity_handler - self.room_member_handler = hs.get_handlers().room_member_handler + self.room_member_handler = hs.get_room_member_handler() self.device_handler = hs.get_device_handler() self.macaroon_gen = hs.get_macaroon_generator() @@ -305,31 +316,67 @@ class RegisterRestServlet(RestServlet): if 'x_show_msisdn' in body and body['x_show_msisdn']: show_msisdn = True + # FIXME: need a better error than "no auth flow found" for scenarios + # where we required 3PID for registration but the user didn't give one + require_email = 'email' in self.hs.config.registrations_require_3pid + require_msisdn = 'msisdn' in self.hs.config.registrations_require_3pid + + flows = [] if self.hs.config.enable_registration_captcha: - flows = [ - [LoginType.RECAPTCHA], - [LoginType.EMAIL_IDENTITY, LoginType.RECAPTCHA], - ] + # only support 3PIDless registration if no 3PIDs are required + if not require_email and not require_msisdn: + flows.extend([[LoginType.RECAPTCHA]]) + # only support the email-only flow if we don't require MSISDN 3PIDs + if not require_msisdn: + flows.extend([[LoginType.EMAIL_IDENTITY, LoginType.RECAPTCHA]]) + if show_msisdn: + # only support the MSISDN-only flow if we don't require email 3PIDs + if not require_email: + flows.extend([[LoginType.MSISDN, LoginType.RECAPTCHA]]) + # always let users provide both MSISDN & email flows.extend([ - [LoginType.MSISDN, LoginType.RECAPTCHA], [LoginType.MSISDN, LoginType.EMAIL_IDENTITY, LoginType.RECAPTCHA], ]) else: - flows = [ - [LoginType.DUMMY], - [LoginType.EMAIL_IDENTITY], - ] + # only support 3PIDless registration if no 3PIDs are required + if not require_email and not require_msisdn: + flows.extend([[LoginType.DUMMY]]) + # only support the email-only flow if we don't require MSISDN 3PIDs + if not require_msisdn: + flows.extend([[LoginType.EMAIL_IDENTITY]]) + if show_msisdn: + # only support the MSISDN-only flow if we don't require email 3PIDs + if not require_email or require_msisdn: + flows.extend([[LoginType.MSISDN]]) + # always let users provide both MSISDN & email flows.extend([ - [LoginType.MSISDN], - [LoginType.MSISDN, LoginType.EMAIL_IDENTITY], + [LoginType.MSISDN, LoginType.EMAIL_IDENTITY] ]) auth_result, params, session_id = yield self.auth_handler.check_auth( flows, body, self.hs.get_ip_from_request(request) ) + # Check that we're not trying to register a denied 3pid. + # + # the user-facing checks will probably already have happened in + # /register/email/requestToken when we requested a 3pid, but that's not + # guaranteed. + + if auth_result: + for login_type in [LoginType.EMAIL_IDENTITY, LoginType.MSISDN]: + if login_type in auth_result: + medium = auth_result[login_type]['medium'] + address = auth_result[login_type]['address'] + + if not check_3pid_allowed(self.hs, medium, address): + raise SynapseError( + 403, "Third party identifier is not allowed", + Codes.THREEPID_DENIED, + ) + if registered_user_id is not None: logger.info( "Already registered user ID %r for this session", diff --git a/synapse/rest/key/v2/remote_key_resource.py b/synapse/rest/key/v2/remote_key_resource.py index cc2842aa72..17e6079cba 100644 --- a/synapse/rest/key/v2/remote_key_resource.py +++ b/synapse/rest/key/v2/remote_key_resource.py @@ -93,6 +93,7 @@ class RemoteKey(Resource): self.store = hs.get_datastore() self.version_string = hs.version_string self.clock = hs.get_clock() + self.federation_domain_whitelist = hs.config.federation_domain_whitelist def render_GET(self, request): self.async_render_GET(request) @@ -137,6 +138,13 @@ class RemoteKey(Resource): logger.info("Handling query for keys %r", query) store_queries = [] for server_name, key_ids in query.items(): + if ( + self.federation_domain_whitelist is not None and + server_name not in self.federation_domain_whitelist + ): + logger.debug("Federation denied with %s", server_name) + continue + if not key_ids: key_ids = (None,) for key_id in key_ids: diff --git a/synapse/rest/media/v1/_base.py b/synapse/rest/media/v1/_base.py index 95fa95fce3..e7ac01da01 100644 --- a/synapse/rest/media/v1/_base.py +++ b/synapse/rest/media/v1/_base.py @@ -70,38 +70,11 @@ def respond_with_file(request, media_type, file_path, logger.debug("Responding with %r", file_path) if os.path.isfile(file_path): - request.setHeader(b"Content-Type", media_type.encode("UTF-8")) - if upload_name: - if is_ascii(upload_name): - request.setHeader( - b"Content-Disposition", - b"inline; filename=%s" % ( - urllib.quote(upload_name.encode("utf-8")), - ), - ) - else: - request.setHeader( - b"Content-Disposition", - b"inline; filename*=utf-8''%s" % ( - urllib.quote(upload_name.encode("utf-8")), - ), - ) - - # cache for at least a day. - # XXX: we might want to turn this off for data we don't want to - # recommend caching as it's sensitive or private - or at least - # select private. don't bother setting Expires as all our - # clients are smart enough to be happy with Cache-Control - request.setHeader( - b"Cache-Control", b"public,max-age=86400,s-maxage=86400" - ) if file_size is None: stat = os.stat(file_path) file_size = stat.st_size - request.setHeader( - b"Content-Length", b"%d" % (file_size,) - ) + add_file_headers(request, media_type, file_size, upload_name) with open(file_path, "rb") as f: yield logcontext.make_deferred_yieldable( @@ -111,3 +84,118 @@ def respond_with_file(request, media_type, file_path, finish_request(request) else: respond_404(request) + + +def add_file_headers(request, media_type, file_size, upload_name): + """Adds the correct response headers in preparation for responding with the + media. + + Args: + request (twisted.web.http.Request) + media_type (str): The media/content type. + file_size (int): Size in bytes of the media, if known. + upload_name (str): The name of the requested file, if any. + """ + request.setHeader(b"Content-Type", media_type.encode("UTF-8")) + if upload_name: + if is_ascii(upload_name): + request.setHeader( + b"Content-Disposition", + b"inline; filename=%s" % ( + urllib.quote(upload_name.encode("utf-8")), + ), + ) + else: + request.setHeader( + b"Content-Disposition", + b"inline; filename*=utf-8''%s" % ( + urllib.quote(upload_name.encode("utf-8")), + ), + ) + + # cache for at least a day. + # XXX: we might want to turn this off for data we don't want to + # recommend caching as it's sensitive or private - or at least + # select private. don't bother setting Expires as all our + # clients are smart enough to be happy with Cache-Control + request.setHeader( + b"Cache-Control", b"public,max-age=86400,s-maxage=86400" + ) + + request.setHeader( + b"Content-Length", b"%d" % (file_size,) + ) + + +@defer.inlineCallbacks +def respond_with_responder(request, responder, media_type, file_size, upload_name=None): + """Responds to the request with given responder. If responder is None then + returns 404. + + Args: + request (twisted.web.http.Request) + responder (Responder|None) + media_type (str): The media/content type. + file_size (int|None): Size in bytes of the media. If not known it should be None + upload_name (str|None): The name of the requested file, if any. + """ + if not responder: + respond_404(request) + return + + add_file_headers(request, media_type, file_size, upload_name) + with responder: + yield responder.write_to_consumer(request) + finish_request(request) + + +class Responder(object): + """Represents a response that can be streamed to the requester. + + Responder is a context manager which *must* be used, so that any resources + held can be cleaned up. + """ + def write_to_consumer(self, consumer): + """Stream response into consumer + + Args: + consumer (IConsumer) + + Returns: + Deferred: Resolves once the response has finished being written + """ + pass + + def __enter__(self): + pass + + def __exit__(self, exc_type, exc_val, exc_tb): + pass + + +class FileInfo(object): + """Details about a requested/uploaded file. + + Attributes: + server_name (str): The server name where the media originated from, + or None if local. + file_id (str): The local ID of the file. For local files this is the + same as the media_id + url_cache (bool): If the file is for the url preview cache + thumbnail (bool): Whether the file is a thumbnail or not. + thumbnail_width (int) + thumbnail_height (int) + thumbnail_method (str) + thumbnail_type (str): Content type of thumbnail, e.g. image/png + """ + def __init__(self, server_name, file_id, url_cache=False, + thumbnail=False, thumbnail_width=None, thumbnail_height=None, + thumbnail_method=None, thumbnail_type=None): + self.server_name = server_name + self.file_id = file_id + self.url_cache = url_cache + self.thumbnail = thumbnail + self.thumbnail_width = thumbnail_width + self.thumbnail_height = thumbnail_height + self.thumbnail_method = thumbnail_method + self.thumbnail_type = thumbnail_type diff --git a/synapse/rest/media/v1/download_resource.py b/synapse/rest/media/v1/download_resource.py index 6879249c8a..fe7e17596f 100644 --- a/synapse/rest/media/v1/download_resource.py +++ b/synapse/rest/media/v1/download_resource.py @@ -14,7 +14,7 @@ # limitations under the License. import synapse.http.servlet -from ._base import parse_media_id, respond_with_file, respond_404 +from ._base import parse_media_id, respond_404 from twisted.web.resource import Resource from synapse.http.server import request_handler, set_cors_headers @@ -32,12 +32,12 @@ class DownloadResource(Resource): def __init__(self, hs, media_repo): Resource.__init__(self) - self.filepaths = media_repo.filepaths self.media_repo = media_repo self.server_name = hs.hostname - self.store = hs.get_datastore() - self.version_string = hs.version_string + + # Both of these are expected by @request_handler() self.clock = hs.get_clock() + self.version_string = hs.version_string def render_GET(self, request): self._async_render_GET(request) @@ -57,59 +57,16 @@ class DownloadResource(Resource): ) server_name, media_id, name = parse_media_id(request) if server_name == self.server_name: - yield self._respond_local_file(request, media_id, name) + yield self.media_repo.get_local_media(request, media_id, name) else: - yield self._respond_remote_file( - request, server_name, media_id, name - ) - - @defer.inlineCallbacks - def _respond_local_file(self, request, media_id, name): - media_info = yield self.store.get_local_media(media_id) - if not media_info or media_info["quarantined_by"]: - respond_404(request) - return - - media_type = media_info["media_type"] - media_length = media_info["media_length"] - upload_name = name if name else media_info["upload_name"] - if media_info["url_cache"]: - # TODO: Check the file still exists, if it doesn't we can redownload - # it from the url `media_info["url_cache"]` - file_path = self.filepaths.url_cache_filepath(media_id) - else: - file_path = self.filepaths.local_media_filepath(media_id) - - yield respond_with_file( - request, media_type, file_path, media_length, - upload_name=upload_name, - ) - - @defer.inlineCallbacks - def _respond_remote_file(self, request, server_name, media_id, name): - # don't forward requests for remote media if allow_remote is false - allow_remote = synapse.http.servlet.parse_boolean( - request, "allow_remote", default=True) - if not allow_remote: - logger.info( - "Rejecting request for remote media %s/%s due to allow_remote", - server_name, media_id, - ) - respond_404(request) - return - - media_info = yield self.media_repo.get_remote_media(server_name, media_id) - - media_type = media_info["media_type"] - media_length = media_info["media_length"] - filesystem_id = media_info["filesystem_id"] - upload_name = name if name else media_info["upload_name"] - - file_path = self.filepaths.remote_media_filepath( - server_name, filesystem_id - ) - - yield respond_with_file( - request, media_type, file_path, media_length, - upload_name=upload_name, - ) + allow_remote = synapse.http.servlet.parse_boolean( + request, "allow_remote", default=True) + if not allow_remote: + logger.info( + "Rejecting request for remote media %s/%s due to allow_remote", + server_name, media_id, + ) + respond_404(request) + return + + yield self.media_repo.get_remote_media(request, server_name, media_id, name) diff --git a/synapse/rest/media/v1/media_repository.py b/synapse/rest/media/v1/media_repository.py index eed9056a2f..bb79599379 100644 --- a/synapse/rest/media/v1/media_repository.py +++ b/synapse/rest/media/v1/media_repository.py @@ -1,5 +1,6 @@ # -*- coding: utf-8 -*- # Copyright 2014-2016 OpenMarket Ltd +# Copyright 2018 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. @@ -18,6 +19,7 @@ import twisted.internet.error import twisted.web.http from twisted.web.resource import Resource +from ._base import respond_404, FileInfo, respond_with_responder from .upload_resource import UploadResource from .download_resource import DownloadResource from .thumbnail_resource import ThumbnailResource @@ -25,15 +27,18 @@ from .identicon_resource import IdenticonResource from .preview_url_resource import PreviewUrlResource from .filepath import MediaFilePaths from .thumbnailer import Thumbnailer +from .storage_provider import StorageProviderWrapper +from .media_storage import MediaStorage from synapse.http.matrixfederationclient import MatrixFederationHttpClient from synapse.util.stringutils import random_string -from synapse.api.errors import SynapseError, HttpResponseException, \ - NotFoundError +from synapse.api.errors import ( + SynapseError, HttpResponseException, NotFoundError, FederationDeniedError, +) from synapse.util.async import Linearizer from synapse.util.stringutils import is_ascii -from synapse.util.logcontext import make_deferred_yieldable, preserve_fn +from synapse.util.logcontext import make_deferred_yieldable from synapse.util.retryutils import NotRetryingDestination import os @@ -47,7 +52,7 @@ import urlparse logger = logging.getLogger(__name__) -UPDATE_RECENTLY_ACCESSED_REMOTES_TS = 60 * 1000 +UPDATE_RECENTLY_ACCESSED_TS = 60 * 1000 class MediaRepository(object): @@ -63,96 +68,62 @@ class MediaRepository(object): self.primary_base_path = hs.config.media_store_path self.filepaths = MediaFilePaths(self.primary_base_path) - self.backup_base_path = hs.config.backup_media_store_path - - self.synchronous_backup_media_store = hs.config.synchronous_backup_media_store - self.dynamic_thumbnails = hs.config.dynamic_thumbnails self.thumbnail_requirements = hs.config.thumbnail_requirements self.remote_media_linearizer = Linearizer(name="media_remote") self.recently_accessed_remotes = set() + self.recently_accessed_locals = set() - self.clock.looping_call( - self._update_recently_accessed_remotes, - UPDATE_RECENTLY_ACCESSED_REMOTES_TS - ) + self.federation_domain_whitelist = hs.config.federation_domain_whitelist - @defer.inlineCallbacks - def _update_recently_accessed_remotes(self): - media = self.recently_accessed_remotes - self.recently_accessed_remotes = set() - - yield self.store.update_cached_last_access_time( - media, self.clock.time_msec() - ) + # List of StorageProviders where we should search for media and + # potentially upload to. + storage_providers = [] - @staticmethod - def _makedirs(filepath): - dirname = os.path.dirname(filepath) - if not os.path.exists(dirname): - os.makedirs(dirname) + for clz, provider_config, wrapper_config in hs.config.media_storage_providers: + backend = clz(hs, provider_config) + provider = StorageProviderWrapper( + backend, + store_local=wrapper_config.store_local, + store_remote=wrapper_config.store_remote, + store_synchronous=wrapper_config.store_synchronous, + ) + storage_providers.append(provider) - @staticmethod - def _write_file_synchronously(source, fname): - """Write `source` to the path `fname` synchronously. Should be called - from a thread. + self.media_storage = MediaStorage( + self.primary_base_path, self.filepaths, storage_providers, + ) - Args: - source: A file like object to be written - fname (str): Path to write to - """ - MediaRepository._makedirs(fname) - source.seek(0) # Ensure we read from the start of the file - with open(fname, "wb") as f: - shutil.copyfileobj(source, f) + self.clock.looping_call( + self._update_recently_accessed, + UPDATE_RECENTLY_ACCESSED_TS, + ) @defer.inlineCallbacks - def write_to_file_and_backup(self, source, path): - """Write `source` to the on disk media store, and also the backup store - if configured. - - Args: - source: A file like object that should be written - path (str): Relative path to write file to - - Returns: - Deferred[str]: the file path written to in the primary media store - """ - fname = os.path.join(self.primary_base_path, path) - - # Write to the main repository - yield make_deferred_yieldable(threads.deferToThread( - self._write_file_synchronously, source, fname, - )) + def _update_recently_accessed(self): + remote_media = self.recently_accessed_remotes + self.recently_accessed_remotes = set() - # Write to backup repository - yield self.copy_to_backup(path) + local_media = self.recently_accessed_locals + self.recently_accessed_locals = set() - defer.returnValue(fname) + yield self.store.update_cached_last_access_time( + local_media, remote_media, self.clock.time_msec() + ) - @defer.inlineCallbacks - def copy_to_backup(self, path): - """Copy a file from the primary to backup media store, if configured. + def mark_recently_accessed(self, server_name, media_id): + """Mark the given media as recently accessed. Args: - path(str): Relative path to write file to + server_name (str|None): Origin server of media, or None if local + media_id (str): The media ID of the content """ - if self.backup_base_path: - primary_fname = os.path.join(self.primary_base_path, path) - backup_fname = os.path.join(self.backup_base_path, path) - - # We can either wait for successful writing to the backup repository - # or write in the background and immediately return - if self.synchronous_backup_media_store: - yield make_deferred_yieldable(threads.deferToThread( - shutil.copyfile, primary_fname, backup_fname, - )) - else: - preserve_fn(threads.deferToThread)( - shutil.copyfile, primary_fname, backup_fname, - ) + if server_name: + self.recently_accessed_remotes.add((server_name, media_id)) + else: + self.recently_accessed_locals.add(media_id) @defer.inlineCallbacks def create_content(self, media_type, upload_name, content, content_length, @@ -171,10 +142,13 @@ class MediaRepository(object): """ media_id = random_string(24) - fname = yield self.write_to_file_and_backup( - content, self.filepaths.local_media_filepath_rel(media_id) + file_info = FileInfo( + server_name=None, + file_id=media_id, ) + fname = yield self.media_storage.store_file(content, file_info) + logger.info("Stored local media in file %r", fname) yield self.store.store_local_media( @@ -185,134 +159,275 @@ class MediaRepository(object): media_length=content_length, user_id=auth_user, ) - media_info = { - "media_type": media_type, - "media_length": content_length, - } - yield self._generate_thumbnails(None, media_id, media_info) + yield self._generate_thumbnails( + None, media_id, media_id, media_type, + ) defer.returnValue("mxc://%s/%s" % (self.server_name, media_id)) @defer.inlineCallbacks - def get_remote_media(self, server_name, media_id): + def get_local_media(self, request, media_id, name): + """Responds to reqests for local media, if exists, or returns 404. + + Args: + request(twisted.web.http.Request) + media_id (str): The media ID of the content. (This is the same as + the file_id for local content.) + name (str|None): Optional name that, if specified, will be used as + the filename in the Content-Disposition header of the response. + + Returns: + Deferred: Resolves once a response has successfully been written + to request + """ + media_info = yield self.store.get_local_media(media_id) + if not media_info or media_info["quarantined_by"]: + respond_404(request) + return + + self.mark_recently_accessed(None, media_id) + + media_type = media_info["media_type"] + media_length = media_info["media_length"] + upload_name = name if name else media_info["upload_name"] + url_cache = media_info["url_cache"] + + file_info = FileInfo( + None, media_id, + url_cache=url_cache, + ) + + responder = yield self.media_storage.fetch_media(file_info) + yield respond_with_responder( + request, responder, media_type, media_length, upload_name, + ) + + @defer.inlineCallbacks + def get_remote_media(self, request, server_name, media_id, name): + """Respond to requests for remote media. + + Args: + request(twisted.web.http.Request) + server_name (str): Remote server_name where the media originated. + media_id (str): The media ID of the content (as defined by the + remote server). + name (str|None): Optional name that, if specified, will be used as + the filename in the Content-Disposition header of the response. + + Returns: + Deferred: Resolves once a response has successfully been written + to request + """ + if ( + self.federation_domain_whitelist is not None and + server_name not in self.federation_domain_whitelist + ): + raise FederationDeniedError(server_name) + + self.mark_recently_accessed(server_name, media_id) + + # We linearize here to ensure that we don't try and download remote + # media multiple times concurrently + key = (server_name, media_id) + with (yield self.remote_media_linearizer.queue(key)): + responder, media_info = yield self._get_remote_media_impl( + server_name, media_id, + ) + + # We deliberately stream the file outside the lock + if responder: + media_type = media_info["media_type"] + media_length = media_info["media_length"] + upload_name = name if name else media_info["upload_name"] + yield respond_with_responder( + request, responder, media_type, media_length, upload_name, + ) + else: + respond_404(request) + + @defer.inlineCallbacks + def get_remote_media_info(self, server_name, media_id): + """Gets the media info associated with the remote file, downloading + if necessary. + + Args: + server_name (str): Remote server_name where the media originated. + media_id (str): The media ID of the content (as defined by the + remote server). + + Returns: + Deferred[dict]: The media_info of the file + """ + if ( + self.federation_domain_whitelist is not None and + server_name not in self.federation_domain_whitelist + ): + raise FederationDeniedError(server_name) + + # We linearize here to ensure that we don't try and download remote + # media multiple times concurrently key = (server_name, media_id) with (yield self.remote_media_linearizer.queue(key)): - media_info = yield self._get_remote_media_impl(server_name, media_id) + responder, media_info = yield self._get_remote_media_impl( + server_name, media_id, + ) + + # Ensure we actually use the responder so that it releases resources + if responder: + with responder: + pass + defer.returnValue(media_info) @defer.inlineCallbacks def _get_remote_media_impl(self, server_name, media_id): + """Looks for media in local cache, if not there then attempt to + download from remote server. + + Args: + server_name (str): Remote server_name where the media originated. + media_id (str): The media ID of the content (as defined by the + remote server). + + Returns: + Deferred[(Responder, media_info)] + """ media_info = yield self.store.get_cached_remote_media( server_name, media_id ) - if not media_info: - media_info = yield self._download_remote_file( - server_name, media_id - ) - elif media_info["quarantined_by"]: - raise NotFoundError() + + # file_id is the ID we use to track the file locally. If we've already + # seen the file then reuse the existing ID, otherwise genereate a new + # one. + if media_info: + file_id = media_info["filesystem_id"] else: - self.recently_accessed_remotes.add((server_name, media_id)) - yield self.store.update_cached_last_access_time( - [(server_name, media_id)], self.clock.time_msec() - ) - defer.returnValue(media_info) + file_id = random_string(24) + + file_info = FileInfo(server_name, file_id) + + # If we have an entry in the DB, try and look for it + if media_info: + if media_info["quarantined_by"]: + logger.info("Media is quarantined") + raise NotFoundError() + + responder = yield self.media_storage.fetch_media(file_info) + if responder: + defer.returnValue((responder, media_info)) + + # Failed to find the file anywhere, lets download it. + + media_info = yield self._download_remote_file( + server_name, media_id, file_id + ) + + responder = yield self.media_storage.fetch_media(file_info) + defer.returnValue((responder, media_info)) @defer.inlineCallbacks - def _download_remote_file(self, server_name, media_id): - file_id = random_string(24) + def _download_remote_file(self, server_name, media_id, file_id): + """Attempt to download the remote file from the given server name, + using the given file_id as the local id. + + Args: + server_name (str): Originating server + media_id (str): The media ID of the content (as defined by the + remote server). This is different than the file_id, which is + locally generated. + file_id (str): Local file ID + + Returns: + Deferred[MediaInfo] + """ - fpath = self.filepaths.remote_media_filepath_rel( - server_name, file_id + file_info = FileInfo( + server_name=server_name, + file_id=file_id, ) - fname = os.path.join(self.primary_base_path, fpath) - self._makedirs(fname) - try: - with open(fname, "wb") as f: - request_path = "/".join(( - "/_matrix/media/v1/download", server_name, media_id, - )) + with self.media_storage.store_into_file(file_info) as (f, fname, finish): + request_path = "/".join(( + "/_matrix/media/v1/download", server_name, media_id, + )) + try: + length, headers = yield self.client.get_file( + server_name, request_path, output_stream=f, + max_size=self.max_upload_size, args={ + # tell the remote server to 404 if it doesn't + # recognise the server_name, to make sure we don't + # end up with a routing loop. + "allow_remote": "false", + } + ) + except twisted.internet.error.DNSLookupError as e: + logger.warn("HTTP error fetching remote media %s/%s: %r", + server_name, media_id, e) + raise NotFoundError() + + except HttpResponseException as e: + logger.warn("HTTP error fetching remote media %s/%s: %s", + server_name, media_id, e.response) + if e.code == twisted.web.http.NOT_FOUND: + raise SynapseError.from_http_response_exception(e) + raise SynapseError(502, "Failed to fetch remote media") + + except SynapseError: + logger.exception("Failed to fetch remote media %s/%s", + server_name, media_id) + raise + except NotRetryingDestination: + logger.warn("Not retrying destination %r", server_name) + raise SynapseError(502, "Failed to fetch remote media") + except Exception: + logger.exception("Failed to fetch remote media %s/%s", + server_name, media_id) + raise SynapseError(502, "Failed to fetch remote media") + + yield finish() + + media_type = headers["Content-Type"][0] + + time_now_ms = self.clock.time_msec() + + content_disposition = headers.get("Content-Disposition", None) + if content_disposition: + _, params = cgi.parse_header(content_disposition[0],) + upload_name = None + + # First check if there is a valid UTF-8 filename + upload_name_utf8 = params.get("filename*", None) + if upload_name_utf8: + if upload_name_utf8.lower().startswith("utf-8''"): + upload_name = upload_name_utf8[7:] + + # If there isn't check for an ascii name. + if not upload_name: + upload_name_ascii = params.get("filename", None) + if upload_name_ascii and is_ascii(upload_name_ascii): + upload_name = upload_name_ascii + + if upload_name: + upload_name = urlparse.unquote(upload_name) try: - length, headers = yield self.client.get_file( - server_name, request_path, output_stream=f, - max_size=self.max_upload_size, args={ - # tell the remote server to 404 if it doesn't - # recognise the server_name, to make sure we don't - # end up with a routing loop. - "allow_remote": "false", - } - ) - except twisted.internet.error.DNSLookupError as e: - logger.warn("HTTP error fetching remote media %s/%s: %r", - server_name, media_id, e) - raise NotFoundError() - - except HttpResponseException as e: - logger.warn("HTTP error fetching remote media %s/%s: %s", - server_name, media_id, e.response) - if e.code == twisted.web.http.NOT_FOUND: - raise SynapseError.from_http_response_exception(e) - raise SynapseError(502, "Failed to fetch remote media") - - except SynapseError: - logger.exception("Failed to fetch remote media %s/%s", - server_name, media_id) - raise - except NotRetryingDestination: - logger.warn("Not retrying destination %r", server_name) - raise SynapseError(502, "Failed to fetch remote media") - except Exception: - logger.exception("Failed to fetch remote media %s/%s", - server_name, media_id) - raise SynapseError(502, "Failed to fetch remote media") - - yield self.copy_to_backup(fpath) - - media_type = headers["Content-Type"][0] - time_now_ms = self.clock.time_msec() - - content_disposition = headers.get("Content-Disposition", None) - if content_disposition: - _, params = cgi.parse_header(content_disposition[0],) - upload_name = None - - # First check if there is a valid UTF-8 filename - upload_name_utf8 = params.get("filename*", None) - if upload_name_utf8: - if upload_name_utf8.lower().startswith("utf-8''"): - upload_name = upload_name_utf8[7:] - - # If there isn't check for an ascii name. - if not upload_name: - upload_name_ascii = params.get("filename", None) - if upload_name_ascii and is_ascii(upload_name_ascii): - upload_name = upload_name_ascii - - if upload_name: - upload_name = urlparse.unquote(upload_name) - try: - upload_name = upload_name.decode("utf-8") - except UnicodeDecodeError: - upload_name = None - else: - upload_name = None - - logger.info("Stored remote media in file %r", fname) - - yield self.store.store_cached_remote_media( - origin=server_name, - media_id=media_id, - media_type=media_type, - time_now_ms=self.clock.time_msec(), - upload_name=upload_name, - media_length=length, - filesystem_id=file_id, - ) - except Exception: - os.remove(fname) - raise + upload_name = upload_name.decode("utf-8") + except UnicodeDecodeError: + upload_name = None + else: + upload_name = None + + logger.info("Stored remote media in file %r", fname) + + yield self.store.store_cached_remote_media( + origin=server_name, + media_id=media_id, + media_type=media_type, + time_now_ms=self.clock.time_msec(), + upload_name=upload_name, + media_length=length, + filesystem_id=file_id, + ) media_info = { "media_type": media_type, @@ -323,7 +438,7 @@ class MediaRepository(object): } yield self._generate_thumbnails( - server_name, media_id, media_info + server_name, media_id, file_id, media_type, ) defer.returnValue(media_info) @@ -357,8 +472,10 @@ class MediaRepository(object): @defer.inlineCallbacks def generate_local_exact_thumbnail(self, media_id, t_width, t_height, - t_method, t_type): - input_path = self.filepaths.local_media_filepath(media_id) + t_method, t_type, url_cache): + input_path = yield self.media_storage.ensure_media_is_in_local_cache(FileInfo( + None, media_id, url_cache=url_cache, + )) thumbnailer = Thumbnailer(input_path) t_byte_source = yield make_deferred_yieldable(threads.deferToThread( @@ -368,11 +485,19 @@ class MediaRepository(object): if t_byte_source: try: - output_path = yield self.write_to_file_and_backup( - t_byte_source, - self.filepaths.local_media_thumbnail_rel( - media_id, t_width, t_height, t_type, t_method - ) + file_info = FileInfo( + server_name=None, + file_id=media_id, + url_cache=url_cache, + thumbnail=True, + thumbnail_width=t_width, + thumbnail_height=t_height, + thumbnail_method=t_method, + thumbnail_type=t_type, + ) + + output_path = yield self.media_storage.store_file( + t_byte_source, file_info, ) finally: t_byte_source.close() @@ -390,7 +515,9 @@ class MediaRepository(object): @defer.inlineCallbacks def generate_remote_exact_thumbnail(self, server_name, file_id, media_id, t_width, t_height, t_method, t_type): - input_path = self.filepaths.remote_media_filepath(server_name, file_id) + input_path = yield self.media_storage.ensure_media_is_in_local_cache(FileInfo( + server_name, file_id, url_cache=False, + )) thumbnailer = Thumbnailer(input_path) t_byte_source = yield make_deferred_yieldable(threads.deferToThread( @@ -400,11 +527,18 @@ class MediaRepository(object): if t_byte_source: try: - output_path = yield self.write_to_file_and_backup( - t_byte_source, - self.filepaths.remote_media_thumbnail_rel( - server_name, file_id, t_width, t_height, t_type, t_method - ) + file_info = FileInfo( + server_name=server_name, + file_id=media_id, + thumbnail=True, + thumbnail_width=t_width, + thumbnail_height=t_height, + thumbnail_method=t_method, + thumbnail_type=t_type, + ) + + output_path = yield self.media_storage.store_file( + t_byte_source, file_info, ) finally: t_byte_source.close() @@ -421,31 +555,29 @@ class MediaRepository(object): defer.returnValue(output_path) @defer.inlineCallbacks - def _generate_thumbnails(self, server_name, media_id, media_info, url_cache=False): + def _generate_thumbnails(self, server_name, media_id, file_id, media_type, + url_cache=False): """Generate and store thumbnails for an image. Args: - server_name(str|None): The server name if remote media, else None if local - media_id(str) - media_info(dict) - url_cache(bool): If we are thumbnailing images downloaded for the URL cache, + server_name (str|None): The server name if remote media, else None if local + media_id (str): The media ID of the content. (This is the same as + the file_id for local content) + file_id (str): Local file ID + media_type (str): The content type of the file + url_cache (bool): If we are thumbnailing images downloaded for the URL cache, used exclusively by the url previewer Returns: Deferred[dict]: Dict with "width" and "height" keys of original image """ - media_type = media_info["media_type"] - file_id = media_info.get("filesystem_id") requirements = self._get_thumbnail_requirements(media_type) if not requirements: return - if server_name: - input_path = self.filepaths.remote_media_filepath(server_name, file_id) - elif url_cache: - input_path = self.filepaths.url_cache_filepath(media_id) - else: - input_path = self.filepaths.local_media_filepath(media_id) + input_path = yield self.media_storage.ensure_media_is_in_local_cache(FileInfo( + server_name, file_id, url_cache=url_cache, + )) thumbnailer = Thumbnailer(input_path) m_width = thumbnailer.width @@ -472,20 +604,6 @@ class MediaRepository(object): # Now we generate the thumbnails for each dimension, store it for (t_width, t_height, t_type), t_method in thumbnails.iteritems(): - # Work out the correct file name for thumbnail - if server_name: - file_path = self.filepaths.remote_media_thumbnail_rel( - server_name, file_id, t_width, t_height, t_type, t_method - ) - elif url_cache: - file_path = self.filepaths.url_cache_thumbnail_rel( - media_id, t_width, t_height, t_type, t_method - ) - else: - file_path = self.filepaths.local_media_thumbnail_rel( - media_id, t_width, t_height, t_type, t_method - ) - # Generate the thumbnail if t_method == "crop": t_byte_source = yield make_deferred_yieldable(threads.deferToThread( @@ -505,9 +623,19 @@ class MediaRepository(object): continue try: - # Write to disk - output_path = yield self.write_to_file_and_backup( - t_byte_source, file_path, + file_info = FileInfo( + server_name=server_name, + file_id=file_id, + thumbnail=True, + thumbnail_width=t_width, + thumbnail_height=t_height, + thumbnail_method=t_method, + thumbnail_type=t_type, + url_cache=url_cache, + ) + + output_path = yield self.media_storage.store_file( + t_byte_source, file_info, ) finally: t_byte_source.close() @@ -620,7 +748,11 @@ class MediaRepositoryResource(Resource): self.putChild("upload", UploadResource(hs, media_repo)) self.putChild("download", DownloadResource(hs, media_repo)) - self.putChild("thumbnail", ThumbnailResource(hs, media_repo)) + self.putChild("thumbnail", ThumbnailResource( + hs, media_repo, media_repo.media_storage, + )) self.putChild("identicon", IdenticonResource()) if hs.config.url_preview_enabled: - self.putChild("preview_url", PreviewUrlResource(hs, media_repo)) + self.putChild("preview_url", PreviewUrlResource( + hs, media_repo, media_repo.media_storage, + )) diff --git a/synapse/rest/media/v1/media_storage.py b/synapse/rest/media/v1/media_storage.py new file mode 100644 index 0000000000..83471b3173 --- /dev/null +++ b/synapse/rest/media/v1/media_storage.py @@ -0,0 +1,259 @@ +# -*- coding: utf-8 -*- +# Copyright 2018 New Vecotr Ltd +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +from twisted.internet import defer, threads +from twisted.protocols.basic import FileSender + +from ._base import Responder + +from synapse.util.file_consumer import BackgroundFileConsumer +from synapse.util.logcontext import make_deferred_yieldable + +import contextlib +import os +import logging +import shutil +import sys + + +logger = logging.getLogger(__name__) + + +class MediaStorage(object): + """Responsible for storing/fetching files from local sources. + + Args: + local_media_directory (str): Base path where we store media on disk + filepaths (MediaFilePaths) + storage_providers ([StorageProvider]): List of StorageProvider that are + used to fetch and store files. + """ + + def __init__(self, local_media_directory, filepaths, storage_providers): + self.local_media_directory = local_media_directory + self.filepaths = filepaths + self.storage_providers = storage_providers + + @defer.inlineCallbacks + def store_file(self, source, file_info): + """Write `source` to the on disk media store, and also any other + configured storage providers + + Args: + source: A file like object that should be written + file_info (FileInfo): Info about the file to store + + Returns: + Deferred[str]: the file path written to in the primary media store + """ + + with self.store_into_file(file_info) as (f, fname, finish_cb): + # Write to the main repository + yield make_deferred_yieldable(threads.deferToThread( + _write_file_synchronously, source, f, + )) + yield finish_cb() + + defer.returnValue(fname) + + @contextlib.contextmanager + def store_into_file(self, file_info): + """Context manager used to get a file like object to write into, as + described by file_info. + + Actually yields a 3-tuple (file, fname, finish_cb), where file is a file + like object that can be written to, fname is the absolute path of file + on disk, and finish_cb is a function that returns a Deferred. + + fname can be used to read the contents from after upload, e.g. to + generate thumbnails. + + finish_cb must be called and waited on after the file has been + successfully been written to. Should not be called if there was an + error. + + Args: + file_info (FileInfo): Info about the file to store + + Example: + + with media_storage.store_into_file(info) as (f, fname, finish_cb): + # .. write into f ... + yield finish_cb() + """ + + path = self._file_info_to_path(file_info) + fname = os.path.join(self.local_media_directory, path) + + dirname = os.path.dirname(fname) + if not os.path.exists(dirname): + os.makedirs(dirname) + + finished_called = [False] + + @defer.inlineCallbacks + def finish(): + for provider in self.storage_providers: + yield provider.store_file(path, file_info) + + finished_called[0] = True + + try: + with open(fname, "wb") as f: + yield f, fname, finish + except Exception: + t, v, tb = sys.exc_info() + try: + os.remove(fname) + except Exception: + pass + raise t, v, tb + + if not finished_called: + raise Exception("Finished callback not called") + + @defer.inlineCallbacks + def fetch_media(self, file_info): + """Attempts to fetch media described by file_info from the local cache + and configured storage providers. + + Args: + file_info (FileInfo) + + Returns: + Deferred[Responder|None]: Returns a Responder if the file was found, + otherwise None. + """ + + path = self._file_info_to_path(file_info) + local_path = os.path.join(self.local_media_directory, path) + if os.path.exists(local_path): + defer.returnValue(FileResponder(open(local_path, "rb"))) + + for provider in self.storage_providers: + res = yield provider.fetch(path, file_info) + if res: + defer.returnValue(res) + + defer.returnValue(None) + + @defer.inlineCallbacks + def ensure_media_is_in_local_cache(self, file_info): + """Ensures that the given file is in the local cache. Attempts to + download it from storage providers if it isn't. + + Args: + file_info (FileInfo) + + Returns: + Deferred[str]: Full path to local file + """ + path = self._file_info_to_path(file_info) + local_path = os.path.join(self.local_media_directory, path) + if os.path.exists(local_path): + defer.returnValue(local_path) + + dirname = os.path.dirname(local_path) + if not os.path.exists(dirname): + os.makedirs(dirname) + + for provider in self.storage_providers: + res = yield provider.fetch(path, file_info) + if res: + with res: + consumer = BackgroundFileConsumer(open(local_path, "w")) + yield res.write_to_consumer(consumer) + yield consumer.wait() + defer.returnValue(local_path) + + raise Exception("file could not be found") + + def _file_info_to_path(self, file_info): + """Converts file_info into a relative path. + + The path is suitable for storing files under a directory, e.g. used to + store files on local FS under the base media repository directory. + + Args: + file_info (FileInfo) + + Returns: + str + """ + if file_info.url_cache: + if file_info.thumbnail: + return self.filepaths.url_cache_thumbnail_rel( + media_id=file_info.file_id, + width=file_info.thumbnail_width, + height=file_info.thumbnail_height, + content_type=file_info.thumbnail_type, + method=file_info.thumbnail_method, + ) + return self.filepaths.url_cache_filepath_rel(file_info.file_id) + + if file_info.server_name: + if file_info.thumbnail: + return self.filepaths.remote_media_thumbnail_rel( + server_name=file_info.server_name, + file_id=file_info.file_id, + width=file_info.thumbnail_width, + height=file_info.thumbnail_height, + content_type=file_info.thumbnail_type, + method=file_info.thumbnail_method + ) + return self.filepaths.remote_media_filepath_rel( + file_info.server_name, file_info.file_id, + ) + + if file_info.thumbnail: + return self.filepaths.local_media_thumbnail_rel( + media_id=file_info.file_id, + width=file_info.thumbnail_width, + height=file_info.thumbnail_height, + content_type=file_info.thumbnail_type, + method=file_info.thumbnail_method + ) + return self.filepaths.local_media_filepath_rel( + file_info.file_id, + ) + + +def _write_file_synchronously(source, dest): + """Write `source` to the file like `dest` synchronously. Should be called + from a thread. + + Args: + source: A file like object that's to be written + dest: A file like object to be written to + """ + source.seek(0) # Ensure we read from the start of the file + shutil.copyfileobj(source, dest) + + +class FileResponder(Responder): + """Wraps an open file that can be sent to a request. + + Args: + open_file (file): A file like object to be streamed ot the client, + is closed when finished streaming. + """ + def __init__(self, open_file): + self.open_file = open_file + + def write_to_consumer(self, consumer): + return FileSender().beginFileTransfer(self.open_file, consumer) + + def __exit__(self, exc_type, exc_val, exc_tb): + self.open_file.close() diff --git a/synapse/rest/media/v1/preview_url_resource.py b/synapse/rest/media/v1/preview_url_resource.py index a413cb6226..0fc21540c6 100644 --- a/synapse/rest/media/v1/preview_url_resource.py +++ b/synapse/rest/media/v1/preview_url_resource.py @@ -12,11 +12,26 @@ # 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 cgi +import datetime +import errno +import fnmatch +import itertools +import logging +import os +import re +import shutil +import sys +import traceback +import simplejson as json +import urlparse from twisted.web.server import NOT_DONE_YET from twisted.internet import defer from twisted.web.resource import Resource +from ._base import FileInfo + from synapse.api.errors import ( SynapseError, Codes, ) @@ -31,25 +46,13 @@ from synapse.http.server import ( from synapse.util.async import ObservableDeferred from synapse.util.stringutils import is_ascii -import os -import re -import fnmatch -import cgi -import simplejson as json -import urlparse -import itertools -import datetime -import errno -import shutil - -import logging logger = logging.getLogger(__name__) class PreviewUrlResource(Resource): isLeaf = True - def __init__(self, hs, media_repo): + def __init__(self, hs, media_repo, media_storage): Resource.__init__(self) self.auth = hs.get_auth() @@ -62,6 +65,7 @@ class PreviewUrlResource(Resource): self.client = SpiderHttpClient(hs) self.media_repo = media_repo self.primary_base_path = media_repo.primary_base_path + self.media_storage = media_storage self.url_preview_url_blacklist = hs.config.url_preview_url_blacklist @@ -182,8 +186,10 @@ class PreviewUrlResource(Resource): logger.debug("got media_info of '%s'" % media_info) if _is_media(media_info['media_type']): + file_id = media_info['filesystem_id'] dims = yield self.media_repo._generate_thumbnails( - None, media_info['filesystem_id'], media_info, url_cache=True, + None, file_id, file_id, media_info["media_type"], + url_cache=True, ) og = { @@ -228,8 +234,10 @@ class PreviewUrlResource(Resource): if _is_media(image_info['media_type']): # TODO: make sure we don't choke on white-on-transparent images + file_id = image_info['filesystem_id'] dims = yield self.media_repo._generate_thumbnails( - None, image_info['filesystem_id'], image_info, url_cache=True, + None, file_id, file_id, image_info["media_type"], + url_cache=True, ) if dims: og["og:image:width"] = dims['width'] @@ -273,21 +281,34 @@ class PreviewUrlResource(Resource): file_id = datetime.date.today().isoformat() + '_' + random_string(16) - fpath = self.filepaths.url_cache_filepath_rel(file_id) - fname = os.path.join(self.primary_base_path, fpath) - self.media_repo._makedirs(fname) + file_info = FileInfo( + server_name=None, + file_id=file_id, + url_cache=True, + ) - try: - with open(fname, "wb") as f: + with self.media_storage.store_into_file(file_info) as (f, fname, finish): + try: logger.debug("Trying to get url '%s'" % url) length, headers, uri, code = yield self.client.get_file( url, output_stream=f, max_size=self.max_spider_size, ) + except Exception as e: # FIXME: pass through 404s and other error messages nicely + logger.warn("Error downloading %s: %r", url, e) + raise SynapseError( + 500, "Failed to download content: %s" % ( + traceback.format_exception_only(sys.exc_type, e), + ), + Codes.UNKNOWN, + ) + yield finish() - yield self.media_repo.copy_to_backup(fpath) - - media_type = headers["Content-Type"][0] + try: + if "Content-Type" in headers: + media_type = headers["Content-Type"][0] + else: + media_type = "application/octet-stream" time_now_ms = self.clock.time_msec() content_disposition = headers.get("Content-Disposition", None) @@ -327,11 +348,11 @@ class PreviewUrlResource(Resource): ) except Exception as e: - os.remove(fname) - raise SynapseError( - 500, ("Failed to download content: %s" % e), - Codes.UNKNOWN - ) + logger.error("Error handling downloaded %s: %r", url, e) + # TODO: we really ought to delete the downloaded file in this + # case, since we won't have recorded it in the db, and will + # therefore not expire it. + raise defer.returnValue({ "media_type": media_type, diff --git a/synapse/rest/media/v1/storage_provider.py b/synapse/rest/media/v1/storage_provider.py new file mode 100644 index 0000000000..c188192f2b --- /dev/null +++ b/synapse/rest/media/v1/storage_provider.py @@ -0,0 +1,140 @@ +# -*- coding: utf-8 -*- +# Copyright 2018 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. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +from twisted.internet import defer, threads + +from .media_storage import FileResponder + +from synapse.config._base import Config +from synapse.util.logcontext import preserve_fn + +import logging +import os +import shutil + + +logger = logging.getLogger(__name__) + + +class StorageProvider(object): + """A storage provider is a service that can store uploaded media and + retrieve them. + """ + def store_file(self, path, file_info): + """Store the file described by file_info. The actual contents can be + retrieved by reading the file in file_info.upload_path. + + Args: + path (str): Relative path of file in local cache + file_info (FileInfo) + + Returns: + Deferred + """ + pass + + def fetch(self, path, file_info): + """Attempt to fetch the file described by file_info and stream it + into writer. + + Args: + path (str): Relative path of file in local cache + file_info (FileInfo) + + Returns: + Deferred(Responder): Returns a Responder if the provider has the file, + otherwise returns None. + """ + pass + + +class StorageProviderWrapper(StorageProvider): + """Wraps a storage provider and provides various config options + + Args: + backend (StorageProvider) + store_local (bool): Whether to store new local files or not. + store_synchronous (bool): Whether to wait for file to be successfully + uploaded, or todo the upload in the backgroud. + store_remote (bool): Whether remote media should be uploaded + """ + def __init__(self, backend, store_local, store_synchronous, store_remote): + self.backend = backend + self.store_local = store_local + self.store_synchronous = store_synchronous + self.store_remote = store_remote + + def store_file(self, path, file_info): + if not file_info.server_name and not self.store_local: + return defer.succeed(None) + + if file_info.server_name and not self.store_remote: + return defer.succeed(None) + + if self.store_synchronous: + return self.backend.store_file(path, file_info) + else: + # TODO: Handle errors. + preserve_fn(self.backend.store_file)(path, file_info) + return defer.succeed(None) + + def fetch(self, path, file_info): + return self.backend.fetch(path, file_info) + + +class FileStorageProviderBackend(StorageProvider): + """A storage provider that stores files in a directory on a filesystem. + + Args: + hs (HomeServer) + config: The config returned by `parse_config`. + """ + + def __init__(self, hs, config): + self.cache_directory = hs.config.media_store_path + self.base_directory = config + + def store_file(self, path, file_info): + """See StorageProvider.store_file""" + + primary_fname = os.path.join(self.cache_directory, path) + backup_fname = os.path.join(self.base_directory, path) + + dirname = os.path.dirname(backup_fname) + if not os.path.exists(dirname): + os.makedirs(dirname) + + return threads.deferToThread( + shutil.copyfile, primary_fname, backup_fname, + ) + + def fetch(self, path, file_info): + """See StorageProvider.fetch""" + + backup_fname = os.path.join(self.base_directory, path) + if os.path.isfile(backup_fname): + return FileResponder(open(backup_fname, "rb")) + + @staticmethod + def parse_config(config): + """Called on startup to parse config supplied. This should parse + the config and raise if there is a problem. + + The returned value is passed into the constructor. + + In this case we only care about a single param, the directory, so let's + just pull that out. + """ + return Config.ensure_directory(config["directory"]) diff --git a/synapse/rest/media/v1/thumbnail_resource.py b/synapse/rest/media/v1/thumbnail_resource.py index 68d56b2b10..58ada49711 100644 --- a/synapse/rest/media/v1/thumbnail_resource.py +++ b/synapse/rest/media/v1/thumbnail_resource.py @@ -14,7 +14,10 @@ # limitations under the License. -from ._base import parse_media_id, respond_404, respond_with_file +from ._base import ( + parse_media_id, respond_404, respond_with_file, FileInfo, + respond_with_responder, +) from twisted.web.resource import Resource from synapse.http.servlet import parse_string, parse_integer from synapse.http.server import request_handler, set_cors_headers @@ -30,12 +33,12 @@ logger = logging.getLogger(__name__) class ThumbnailResource(Resource): isLeaf = True - def __init__(self, hs, media_repo): + def __init__(self, hs, media_repo, media_storage): Resource.__init__(self) self.store = hs.get_datastore() - self.filepaths = media_repo.filepaths self.media_repo = media_repo + self.media_storage = media_storage self.dynamic_thumbnails = hs.config.dynamic_thumbnails self.server_name = hs.hostname self.version_string = hs.version_string @@ -64,6 +67,7 @@ class ThumbnailResource(Resource): yield self._respond_local_thumbnail( request, media_id, width, height, method, m_type ) + self.media_repo.mark_recently_accessed(None, media_id) else: if self.dynamic_thumbnails: yield self._select_or_generate_remote_thumbnail( @@ -75,20 +79,20 @@ class ThumbnailResource(Resource): request, server_name, media_id, width, height, method, m_type ) + self.media_repo.mark_recently_accessed(server_name, media_id) @defer.inlineCallbacks def _respond_local_thumbnail(self, request, media_id, width, height, method, m_type): media_info = yield self.store.get_local_media(media_id) - if not media_info or media_info["quarantined_by"]: + if not media_info: + respond_404(request) + return + if media_info["quarantined_by"]: + logger.info("Media is quarantined") respond_404(request) return - - # if media_info["media_type"] == "image/svg+xml": - # file_path = self.filepaths.local_media_filepath(media_id) - # yield respond_with_file(request, media_info["media_type"], file_path) - # return thumbnail_infos = yield self.store.get_local_media_thumbnails(media_id) @@ -96,42 +100,39 @@ class ThumbnailResource(Resource): thumbnail_info = self._select_thumbnail( width, height, method, m_type, thumbnail_infos ) - t_width = thumbnail_info["thumbnail_width"] - t_height = thumbnail_info["thumbnail_height"] - t_type = thumbnail_info["thumbnail_type"] - t_method = thumbnail_info["thumbnail_method"] - - if media_info["url_cache"]: - # TODO: Check the file still exists, if it doesn't we can redownload - # it from the url `media_info["url_cache"]` - file_path = self.filepaths.url_cache_thumbnail( - media_id, t_width, t_height, t_type, t_method, - ) - else: - file_path = self.filepaths.local_media_thumbnail( - media_id, t_width, t_height, t_type, t_method, - ) - yield respond_with_file(request, t_type, file_path) - else: - yield self._respond_default_thumbnail( - request, media_info, width, height, method, m_type, + file_info = FileInfo( + server_name=None, file_id=media_id, + url_cache=media_info["url_cache"], + thumbnail=True, + thumbnail_width=thumbnail_info["thumbnail_width"], + thumbnail_height=thumbnail_info["thumbnail_height"], + thumbnail_type=thumbnail_info["thumbnail_type"], + thumbnail_method=thumbnail_info["thumbnail_method"], ) + t_type = file_info.thumbnail_type + t_length = thumbnail_info["thumbnail_length"] + + responder = yield self.media_storage.fetch_media(file_info) + yield respond_with_responder(request, responder, t_type, t_length) + else: + logger.info("Couldn't find any generated thumbnails") + respond_404(request) + @defer.inlineCallbacks def _select_or_generate_local_thumbnail(self, request, media_id, desired_width, desired_height, desired_method, desired_type): media_info = yield self.store.get_local_media(media_id) - if not media_info or media_info["quarantined_by"]: + if not media_info: + respond_404(request) + return + if media_info["quarantined_by"]: + logger.info("Media is quarantined") respond_404(request) return - - # if media_info["media_type"] == "image/svg+xml": - # file_path = self.filepaths.local_media_filepath(media_id) - # yield respond_with_file(request, media_info["media_type"], file_path) - # return thumbnail_infos = yield self.store.get_local_media_thumbnails(media_id) for info in thumbnail_infos: @@ -141,46 +142,43 @@ class ThumbnailResource(Resource): t_type = info["thumbnail_type"] == desired_type if t_w and t_h and t_method and t_type: - if media_info["url_cache"]: - # TODO: Check the file still exists, if it doesn't we can redownload - # it from the url `media_info["url_cache"]` - file_path = self.filepaths.url_cache_thumbnail( - media_id, desired_width, desired_height, desired_type, - desired_method, - ) - else: - file_path = self.filepaths.local_media_thumbnail( - media_id, desired_width, desired_height, desired_type, - desired_method, - ) - yield respond_with_file(request, desired_type, file_path) - return - - logger.debug("We don't have a local thumbnail of that size. Generating") + file_info = FileInfo( + server_name=None, file_id=media_id, + url_cache=media_info["url_cache"], + thumbnail=True, + thumbnail_width=info["thumbnail_width"], + thumbnail_height=info["thumbnail_height"], + thumbnail_type=info["thumbnail_type"], + thumbnail_method=info["thumbnail_method"], + ) + + t_type = file_info.thumbnail_type + t_length = info["thumbnail_length"] + + responder = yield self.media_storage.fetch_media(file_info) + if responder: + yield respond_with_responder(request, responder, t_type, t_length) + return + + logger.debug("We don't have a thumbnail of that size. Generating") # Okay, so we generate one. file_path = yield self.media_repo.generate_local_exact_thumbnail( - media_id, desired_width, desired_height, desired_method, desired_type + media_id, desired_width, desired_height, desired_method, desired_type, + url_cache=media_info["url_cache"], ) if file_path: yield respond_with_file(request, desired_type, file_path) else: - yield self._respond_default_thumbnail( - request, media_info, desired_width, desired_height, - desired_method, desired_type, - ) + logger.warn("Failed to generate thumbnail") + respond_404(request) @defer.inlineCallbacks def _select_or_generate_remote_thumbnail(self, request, server_name, media_id, desired_width, desired_height, desired_method, desired_type): - media_info = yield self.media_repo.get_remote_media(server_name, media_id) - - # if media_info["media_type"] == "image/svg+xml": - # file_path = self.filepaths.remote_media_filepath(server_name, media_id) - # yield respond_with_file(request, media_info["media_type"], file_path) - # return + media_info = yield self.media_repo.get_remote_media_info(server_name, media_id) thumbnail_infos = yield self.store.get_remote_media_thumbnails( server_name, media_id, @@ -195,14 +193,24 @@ class ThumbnailResource(Resource): t_type = info["thumbnail_type"] == desired_type if t_w and t_h and t_method and t_type: - file_path = self.filepaths.remote_media_thumbnail( - server_name, file_id, desired_width, desired_height, - desired_type, desired_method, + file_info = FileInfo( + server_name=server_name, file_id=media_info["filesystem_id"], + thumbnail=True, + thumbnail_width=info["thumbnail_width"], + thumbnail_height=info["thumbnail_height"], + thumbnail_type=info["thumbnail_type"], + thumbnail_method=info["thumbnail_method"], ) - yield respond_with_file(request, desired_type, file_path) - return - logger.debug("We don't have a local thumbnail of that size. Generating") + t_type = file_info.thumbnail_type + t_length = info["thumbnail_length"] + + responder = yield self.media_storage.fetch_media(file_info) + if responder: + yield respond_with_responder(request, responder, t_type, t_length) + return + + logger.debug("We don't have a thumbnail of that size. Generating") # Okay, so we generate one. file_path = yield self.media_repo.generate_remote_exact_thumbnail( @@ -213,22 +221,16 @@ class ThumbnailResource(Resource): if file_path: yield respond_with_file(request, desired_type, file_path) else: - yield self._respond_default_thumbnail( - request, media_info, desired_width, desired_height, - desired_method, desired_type, - ) + logger.warn("Failed to generate thumbnail") + respond_404(request) @defer.inlineCallbacks def _respond_remote_thumbnail(self, request, server_name, media_id, width, height, method, m_type): # TODO: Don't download the whole remote file - # We should proxy the thumbnail from the remote server instead. - media_info = yield self.media_repo.get_remote_media(server_name, media_id) - - # if media_info["media_type"] == "image/svg+xml": - # file_path = self.filepaths.remote_media_filepath(server_name, media_id) - # yield respond_with_file(request, media_info["media_type"], file_path) - # return + # We should proxy the thumbnail from the remote server instead of + # downloading the remote file and generating our own thumbnails. + media_info = yield self.media_repo.get_remote_media_info(server_name, media_id) thumbnail_infos = yield self.store.get_remote_media_thumbnails( server_name, media_id, @@ -238,59 +240,23 @@ class ThumbnailResource(Resource): thumbnail_info = self._select_thumbnail( width, height, method, m_type, thumbnail_infos ) - t_width = thumbnail_info["thumbnail_width"] - t_height = thumbnail_info["thumbnail_height"] - t_type = thumbnail_info["thumbnail_type"] - t_method = thumbnail_info["thumbnail_method"] - file_id = thumbnail_info["filesystem_id"] + file_info = FileInfo( + server_name=server_name, file_id=media_info["filesystem_id"], + thumbnail=True, + thumbnail_width=thumbnail_info["thumbnail_width"], + thumbnail_height=thumbnail_info["thumbnail_height"], + thumbnail_type=thumbnail_info["thumbnail_type"], + thumbnail_method=thumbnail_info["thumbnail_method"], + ) + + t_type = file_info.thumbnail_type t_length = thumbnail_info["thumbnail_length"] - file_path = self.filepaths.remote_media_thumbnail( - server_name, file_id, t_width, t_height, t_type, t_method, - ) - yield respond_with_file(request, t_type, file_path, t_length) + responder = yield self.media_storage.fetch_media(file_info) + yield respond_with_responder(request, responder, t_type, t_length) else: - yield self._respond_default_thumbnail( - request, media_info, width, height, method, m_type, - ) - - @defer.inlineCallbacks - def _respond_default_thumbnail(self, request, media_info, width, height, - method, m_type): - # XXX: how is this meant to work? store.get_default_thumbnails - # appears to always return [] so won't this always 404? - media_type = media_info["media_type"] - top_level_type = media_type.split("/")[0] - sub_type = media_type.split("/")[-1].split(";")[0] - thumbnail_infos = yield self.store.get_default_thumbnails( - top_level_type, sub_type, - ) - if not thumbnail_infos: - thumbnail_infos = yield self.store.get_default_thumbnails( - top_level_type, "_default", - ) - if not thumbnail_infos: - thumbnail_infos = yield self.store.get_default_thumbnails( - "_default", "_default", - ) - if not thumbnail_infos: + logger.info("Failed to find any generated thumbnails") respond_404(request) - return - - thumbnail_info = self._select_thumbnail( - width, height, "crop", m_type, thumbnail_infos - ) - - t_width = thumbnail_info["thumbnail_width"] - t_height = thumbnail_info["thumbnail_height"] - t_type = thumbnail_info["thumbnail_type"] - t_method = thumbnail_info["thumbnail_method"] - t_length = thumbnail_info["thumbnail_length"] - - file_path = self.filepaths.default_thumbnail( - top_level_type, sub_type, t_width, t_height, t_type, t_method, - ) - yield respond_with_file(request, t_type, file_path, t_length) def _select_thumbnail(self, desired_width, desired_height, desired_method, desired_type, thumbnail_infos): diff --git a/synapse/server.py b/synapse/server.py index 99693071b6..cd0c1a51be 100644 --- a/synapse/server.py +++ b/synapse/server.py @@ -32,8 +32,10 @@ from synapse.appservice.scheduler import ApplicationServiceScheduler from synapse.crypto.keyring import Keyring from synapse.events.builder import EventBuilderFactory from synapse.events.spamcheck import SpamChecker -from synapse.federation import initialize_http_replication +from synapse.federation.federation_client import FederationClient +from synapse.federation.federation_server import FederationServer from synapse.federation.send_queue import FederationRemoteSendQueue +from synapse.federation.federation_server import FederationHandlerRegistry from synapse.federation.transport.client import TransportLayerClient from synapse.federation.transaction_queue import TransactionQueue from synapse.handlers import Handlers @@ -45,6 +47,8 @@ from synapse.handlers.device import DeviceHandler from synapse.handlers.e2e_keys import E2eKeysHandler from synapse.handlers.presence import PresenceHandler from synapse.handlers.room_list import RoomListHandler +from synapse.handlers.room_member import RoomMemberMasterHandler +from synapse.handlers.room_member_worker import RoomMemberWorkerHandler from synapse.handlers.set_password import SetPasswordHandler from synapse.handlers.sync import SyncHandler from synapse.handlers.typing import TypingHandler @@ -55,6 +59,7 @@ from synapse.handlers.read_marker import ReadMarkerHandler from synapse.handlers.user_directory import UserDirectoryHandler from synapse.handlers.groups_local import GroupsLocalHandler from synapse.handlers.profile import ProfileHandler +from synapse.handlers.message import EventCreationHandler from synapse.groups.groups_server import GroupsServerHandler from synapse.groups.attestations import GroupAttestionRenewer, GroupAttestationSigning from synapse.http.client import SimpleHttpClient, InsecureInterceptableContextFactory @@ -66,7 +71,7 @@ from synapse.rest.media.v1.media_repository import ( MediaRepository, MediaRepositoryResource, ) -from synapse.state import StateHandler +from synapse.state import StateHandler, StateResolutionHandler from synapse.storage import DataStore from synapse.streams.events import EventSources from synapse.util import Clock @@ -97,11 +102,13 @@ class HomeServer(object): DEPENDENCIES = [ 'http_client', 'db_pool', - 'replication_layer', + 'federation_client', + 'federation_server', 'handlers', 'v1auth', 'auth', 'state_handler', + 'state_resolution_handler', 'presence_handler', 'sync_handler', 'typing_handler', @@ -117,6 +124,7 @@ class HomeServer(object): 'application_service_handler', 'device_message_handler', 'profile_handler', + 'event_creation_handler', 'deactivate_account_handler', 'set_password_handler', 'notifier', @@ -142,6 +150,8 @@ class HomeServer(object): 'groups_attestation_signing', 'groups_attestation_renewer', 'spam_checker', + 'room_member_handler', + 'federation_registry', ] def __init__(self, hostname, **kwargs): @@ -190,8 +200,11 @@ class HomeServer(object): def get_ratelimiter(self): return self.ratelimiter - def build_replication_layer(self): - return initialize_http_replication(self) + def build_federation_client(self): + return FederationClient(self) + + def build_federation_server(self): + return FederationServer(self) def build_handlers(self): return Handlers(self) @@ -224,6 +237,9 @@ class HomeServer(object): def build_state_handler(self): return StateHandler(self) + def build_state_resolution_handler(self): + return StateResolutionHandler(self) + def build_presence_handler(self): return PresenceHandler(self) @@ -272,6 +288,9 @@ class HomeServer(object): def build_profile_handler(self): return ProfileHandler(self) + def build_event_creation_handler(self): + return EventCreationHandler(self) + def build_deactivate_account_handler(self): return DeactivateAccountHandler(self) @@ -307,6 +326,23 @@ class HomeServer(object): **self.db_config.get("args", {}) ) + def get_db_conn(self, run_new_connection=True): + """Makes a new connection to the database, skipping the db pool + + Returns: + Connection: a connection object implementing the PEP-249 spec + """ + # Any param beginning with cp_ is a parameter for adbapi, and should + # not be passed to the database engine. + db_params = { + k: v for k, v in self.db_config.get("args", {}).items() + if not k.startswith("cp_") + } + db_conn = self.database_engine.module.connect(**db_params) + if run_new_connection: + self.database_engine.on_new_connection(db_conn) + return db_conn + def build_media_repository_resource(self): # build the media repo resource. This indirects through the HomeServer # to ensure that we only have a single instance of @@ -356,6 +392,14 @@ class HomeServer(object): def build_spam_checker(self): return SpamChecker(self) + def build_room_member_handler(self): + if self.config.worker_app: + return RoomMemberWorkerHandler(self) + return RoomMemberMasterHandler(self) + + def build_federation_registry(self): + return FederationHandlerRegistry() + def remove_pusher(self, app_id, push_key, user_id): return self.get_pusherpool().remove_pusher(app_id, push_key, user_id) diff --git a/synapse/server.pyi b/synapse/server.pyi index 41416ef252..c3a9a3847b 100644 --- a/synapse/server.pyi +++ b/synapse/server.pyi @@ -34,6 +34,9 @@ class HomeServer(object): def get_state_handler(self) -> synapse.state.StateHandler: pass + def get_state_resolution_handler(self) -> synapse.state.StateResolutionHandler: + pass + def get_deactivate_account_handler(self) -> synapse.handlers.deactivate_account.DeactivateAccountHandler: pass diff --git a/synapse/state.py b/synapse/state.py index 9e624b4937..932f602508 100644 --- a/synapse/state.py +++ b/synapse/state.py @@ -58,7 +58,11 @@ class _StateCacheEntry(object): __slots__ = ["state", "state_group", "state_id", "prev_group", "delta_ids"] def __init__(self, state, state_group, prev_group=None, delta_ids=None): + # dict[(str, str), str] map from (type, state_key) to event_id self.state = frozendict(state) + + # the ID of a state group if one and only one is involved. + # otherwise, None otherwise? self.state_group = state_group self.prev_group = prev_group @@ -81,31 +85,19 @@ class _StateCacheEntry(object): class StateHandler(object): - """ Responsible for doing state conflict resolution. + """Fetches bits of state from the stores, and does state resolution + where necessary """ def __init__(self, hs): self.clock = hs.get_clock() self.store = hs.get_datastore() self.hs = hs - - # dict of set of event_ids -> _StateCacheEntry. - self._state_cache = None - self.resolve_linearizer = Linearizer(name="state_resolve_lock") + self._state_resolution_handler = hs.get_state_resolution_handler() def start_caching(self): - logger.debug("start_caching") - - self._state_cache = ExpiringCache( - cache_name="state_cache", - clock=self.clock, - max_len=SIZE_OF_CACHE, - expiry_ms=EVICTION_TIMEOUT_SECONDS * 1000, - iterable=True, - reset_expiry_on_get=True, - ) - - self._state_cache.start() + # TODO: remove this shim + self._state_resolution_handler.start_caching() @defer.inlineCallbacks def get_current_state(self, room_id, event_type=None, state_key="", @@ -127,7 +119,7 @@ class StateHandler(object): latest_event_ids = yield self.store.get_latest_event_ids_in_room(room_id) logger.debug("calling resolve_state_groups from get_current_state") - ret = yield self.resolve_state_groups(room_id, latest_event_ids) + ret = yield self.resolve_state_groups_for_events(room_id, latest_event_ids) state = ret.state if event_type: @@ -146,19 +138,27 @@ class StateHandler(object): defer.returnValue(state) @defer.inlineCallbacks - def get_current_state_ids(self, room_id, event_type=None, state_key="", - latest_event_ids=None): + def get_current_state_ids(self, room_id, latest_event_ids=None): + """Get the current state, or the state at a set of events, for a room + + Args: + room_id (str): + + latest_event_ids (iterable[str]|None): if given, the forward + extremities to resolve. If None, we look them up from the + database (via a cache) + + Returns: + Deferred[dict[(str, str), str)]]: the state dict, mapping from + (event_type, state_key) -> event_id + """ if not latest_event_ids: latest_event_ids = yield self.store.get_latest_event_ids_in_room(room_id) logger.debug("calling resolve_state_groups from get_current_state_ids") - ret = yield self.resolve_state_groups(room_id, latest_event_ids) + ret = yield self.resolve_state_groups_for_events(room_id, latest_event_ids) state = ret.state - if event_type: - defer.returnValue(state.get((event_type, state_key))) - return - defer.returnValue(state) @defer.inlineCallbacks @@ -166,7 +166,7 @@ class StateHandler(object): if not latest_event_ids: latest_event_ids = yield self.store.get_latest_event_ids_in_room(room_id) logger.debug("calling resolve_state_groups from get_current_user_in_room") - entry = yield self.resolve_state_groups(room_id, latest_event_ids) + entry = yield self.resolve_state_groups_for_events(room_id, latest_event_ids) joined_users = yield self.store.get_joined_users_from_state(room_id, entry) defer.returnValue(joined_users) @@ -175,7 +175,7 @@ class StateHandler(object): if not latest_event_ids: latest_event_ids = yield self.store.get_latest_event_ids_in_room(room_id) logger.debug("calling resolve_state_groups from get_current_hosts_in_room") - entry = yield self.resolve_state_groups(room_id, latest_event_ids) + entry = yield self.resolve_state_groups_for_events(room_id, latest_event_ids) joined_hosts = yield self.store.get_joined_hosts(room_id, entry) defer.returnValue(joined_hosts) @@ -183,8 +183,15 @@ class StateHandler(object): def compute_event_context(self, event, old_state=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 + 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. Returns: synapse.events.snapshot.EventContext: """ @@ -208,15 +215,22 @@ class StateHandler(object): context.current_state_ids = {} context.prev_state_ids = {} context.prev_state_events = [] - context.state_group = self.store.get_next_state_group() + + # We don't store state for outliers, so we don't generate a state + # froup for it. + context.state_group = None + defer.returnValue(context) 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. + context = EventContext() context.prev_state_ids = { (s.type, s.state_key): s.event_id for s in old_state } - context.state_group = self.store.get_next_state_group() if event.is_state(): key = (event.type, event.state_key) @@ -229,11 +243,19 @@ class StateHandler(object): else: context.current_state_ids = context.prev_state_ids + context.state_group = yield self.store.store_state_group( + event.event_id, + event.room_id, + prev_group=None, + delta_ids=None, + current_state_ids=context.current_state_ids, + ) + context.prev_state_events = [] defer.returnValue(context) logger.debug("calling resolve_state_groups from compute_event_context") - entry = yield self.resolve_state_groups( + entry = yield self.resolve_state_groups_for_events( event.room_id, [e for e, _ in event.prev_events], ) @@ -242,7 +264,8 @@ class StateHandler(object): context = EventContext() context.prev_state_ids = curr_state if event.is_state(): - context.state_group = self.store.get_next_state_group() + # If this is a state event then we need to create a new state + # group for the state after this event. key = (event.type, event.state_key) if key in context.prev_state_ids: @@ -253,38 +276,57 @@ class StateHandler(object): context.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 context.prev_group = entry.state_group context.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. context.prev_group = entry.prev_group context.delta_ids = dict(entry.delta_ids) context.delta_ids[key] = event.event_id + + context.state_group = yield self.store.store_state_group( + event.event_id, + event.room_id, + prev_group=context.prev_group, + delta_ids=context.delta_ids, + current_state_ids=context.current_state_ids, + ) else: + context.current_state_ids = context.prev_state_ids + context.prev_group = entry.prev_group + context.delta_ids = entry.delta_ids + if entry.state_group is None: - entry.state_group = self.store.get_next_state_group() + entry.state_group = yield self.store.store_state_group( + event.event_id, + event.room_id, + prev_group=entry.prev_group, + delta_ids=entry.delta_ids, + current_state_ids=context.current_state_ids, + ) entry.state_id = entry.state_group context.state_group = entry.state_group - context.current_state_ids = context.prev_state_ids - context.prev_group = entry.prev_group - context.delta_ids = entry.delta_ids context.prev_state_events = [] defer.returnValue(context) @defer.inlineCallbacks - @log_function - def resolve_state_groups(self, room_id, event_ids): + def resolve_state_groups_for_events(self, room_id, event_ids): """ Given a list of event_ids this method fetches the state at each event, resolves conflicts between them and returns them. + Args: + room_id (str): + event_ids (list[str]): + Returns: - a Deferred tuple of (`state_group`, `state`, `prev_state`). - `state_group` is the name of a state group if one and only one is - involved. `state` is a map from (type, state_key) to event, and - `prev_state` is a list of event ids. + Deferred[_StateCacheEntry]: resolved state """ logger.debug("resolve_state_groups event_ids %s", event_ids) @@ -295,13 +337,7 @@ class StateHandler(object): room_id, event_ids ) - logger.debug( - "resolve_state_groups state_groups %s", - state_groups_ids.keys() - ) - - group_names = frozenset(state_groups_ids.keys()) - if len(group_names) == 1: + if len(state_groups_ids) == 1: name, state_list = state_groups_ids.items().pop() prev_group, delta_ids = yield self.store.get_state_group_delta(name) @@ -313,6 +349,102 @@ class StateHandler(object): delta_ids=delta_ids, )) + result = yield self._state_resolution_handler.resolve_state_groups( + room_id, state_groups_ids, None, self._state_map_factory, + ) + defer.returnValue(result) + + def _state_map_factory(self, ev_ids): + return self.store.get_events( + ev_ids, get_prev_content=False, check_redacted=False, + ) + + def resolve_events(self, state_sets, event): + logger.info( + "Resolving state for %s with %d groups", event.room_id, len(state_sets) + ) + state_set_ids = [{ + (ev.type, ev.state_key): ev.event_id + for ev in st + } for st in state_sets] + + state_map = { + ev.event_id: ev + for st in state_sets + for ev in st + } + + with Measure(self.clock, "state._resolve_events"): + new_state = resolve_events_with_state_map(state_set_ids, state_map) + + new_state = { + key: state_map[ev_id] for key, ev_id in new_state.items() + } + + return new_state + + +class StateResolutionHandler(object): + """Responsible for doing state conflict resolution. + + Note that the storage layer depends on this handler, so all functions must + be storage-independent. + """ + def __init__(self, hs): + self.clock = hs.get_clock() + + # dict of set of event_ids -> _StateCacheEntry. + self._state_cache = None + self.resolve_linearizer = Linearizer(name="state_resolve_lock") + + def start_caching(self): + logger.debug("start_caching") + + self._state_cache = ExpiringCache( + cache_name="state_cache", + clock=self.clock, + max_len=SIZE_OF_CACHE, + expiry_ms=EVICTION_TIMEOUT_SECONDS * 1000, + iterable=True, + reset_expiry_on_get=True, + ) + + self._state_cache.start() + + @defer.inlineCallbacks + @log_function + def resolve_state_groups( + self, room_id, state_groups_ids, event_map, state_map_factory, + ): + """Resolves conflicts between a set of state groups + + Always generates a new state group (unless we hit the cache), so should + not be called for a single state group + + Args: + room_id (str): room we are resolving for (used for logging) + state_groups_ids (dict[int, dict[(str, str), str]]): + map from state group id to the state in that state group + (where 'state' is a map from state key to event id) + + event_map(dict[str,FrozenEvent]|None): + a dict from event_id to event, for any events that we happen to + have in flight (eg, those currently being persisted). This will be + used as a starting point fof finding the state we need; any missing + events will be requested via state_map_factory. + + If None, all events will be fetched via state_map_factory. + + Returns: + Deferred[_StateCacheEntry]: resolved state + """ + logger.debug( + "resolve_state_groups state_groups %s", + state_groups_ids.keys() + ) + + group_names = frozenset(state_groups_ids.keys()) + with (yield self.resolve_linearizer.queue(group_names)): if self._state_cache is not None: cache = self._state_cache.get(group_names, None) @@ -341,17 +473,20 @@ class StateHandler(object): if conflicted_state: logger.info("Resolving conflicted state for %r", room_id) with Measure(self.clock, "state._resolve_events"): - new_state = yield resolve_events( + new_state = yield resolve_events_with_factory( state_groups_ids.values(), - state_map_factory=lambda ev_ids: self.store.get_events( - ev_ids, get_prev_content=False, check_redacted=False, - ), + event_map=event_map, + state_map_factory=state_map_factory, ) else: new_state = { key: e_ids.pop() for key, e_ids in state.items() } + # if the new state matches any of the input state groups, we can + # use that state group again. Otherwise we will generate a state_id + # which will be used as a cache key for future resolutions, but + # not get persisted. state_group = None new_state_event_ids = frozenset(new_state.values()) for sg, events in state_groups_ids.items(): @@ -388,30 +523,6 @@ class StateHandler(object): defer.returnValue(cache) - def resolve_events(self, state_sets, event): - logger.info( - "Resolving state for %s with %d groups", event.room_id, len(state_sets) - ) - state_set_ids = [{ - (ev.type, ev.state_key): ev.event_id - for ev in st - } for st in state_sets] - - state_map = { - ev.event_id: ev - for st in state_sets - for ev in st - } - - with Measure(self.clock, "state._resolve_events"): - new_state = resolve_events(state_set_ids, state_map) - - new_state = { - key: state_map[ev_id] for key, ev_id in new_state.items() - } - - return new_state - def _ordered_events(events): def key_func(e): @@ -420,19 +531,17 @@ def _ordered_events(events): return sorted(events, key=key_func) -def resolve_events(state_sets, state_map_factory): +def resolve_events_with_state_map(state_sets, state_map): """ Args: state_sets(list): List of dicts of (type, state_key) -> event_id, which are the different state groups to resolve. - state_map_factory(dict|callable): If callable, then will be called - with a list of event_ids that are needed, and should return with - a Deferred of dict of event_id to event. Otherwise, should be - a dict from event_id to event of all events in state_sets. + state_map(dict): a dict from event_id to event, for all events in + state_sets. Returns - dict[(str, str), synapse.events.FrozenEvent] is a map from - (type, state_key) to event. + dict[(str, str), str]: + a map from (type, state_key) to event_id. """ if len(state_sets) == 1: return state_sets[0] @@ -441,13 +550,6 @@ def resolve_events(state_sets, state_map_factory): state_sets, ) - if callable(state_map_factory): - return _resolve_with_state_fac( - unconflicted_state, conflicted_state, state_map_factory - ) - - state_map = state_map_factory - auth_events = _create_auth_events_from_maps( unconflicted_state, conflicted_state, state_map ) @@ -461,6 +563,21 @@ def _seperate(state_sets): """Takes the state_sets and figures out which keys are conflicted and which aren't. i.e., which have multiple different event_ids associated with them in different state sets. + + Args: + state_sets(list[dict[(str, str), str]]): + List of dicts of (type, state_key) -> event_id, which are the + different state groups to resolve. + + Returns: + (dict[(str, str), str], dict[(str, str), set[str]]): + A tuple of (unconflicted_state, conflicted_state), where: + + unconflicted_state is a dict mapping (type, state_key)->event_id + for unconflicted state keys. + + conflicted_state is a dict mapping (type, state_key) to a set of + event ids for conflicted state keys. """ unconflicted_state = dict(state_sets[0]) conflicted_state = {} @@ -491,19 +608,50 @@ def _seperate(state_sets): @defer.inlineCallbacks -def _resolve_with_state_fac(unconflicted_state, conflicted_state, - state_map_factory): +def resolve_events_with_factory(state_sets, event_map, state_map_factory): + """ + Args: + state_sets(list): List of dicts of (type, state_key) -> event_id, + which are the different state groups to resolve. + + event_map(dict[str,FrozenEvent]|None): + a dict from event_id to event, for any events that we happen to + have in flight (eg, those currently being persisted). This will be + used as a starting point fof finding the state we need; any missing + events will be requested via state_map_factory. + + If None, all events will be fetched via state_map_factory. + + state_map_factory(func): will be called + with a list of event_ids that are needed, and should return with + a Deferred of dict of event_id to event. + + Returns + Deferred[dict[(str, str), str]]: + a map from (type, state_key) to event_id. + """ + if len(state_sets) == 1: + defer.returnValue(state_sets[0]) + + unconflicted_state, conflicted_state = _seperate( + state_sets, + ) + needed_events = set( event_id for event_ids in conflicted_state.itervalues() for event_id in event_ids ) + if event_map is not None: + needed_events -= set(event_map.iterkeys()) logger.info("Asking for %d conflicted events", len(needed_events)) # dict[str, FrozenEvent]: a map from state event id to event. Only includes - # the state events which are in conflict. + # the state events which are in conflict (and those in event_map) state_map = yield state_map_factory(needed_events) + if event_map is not None: + state_map.update(event_map) # get the ids of the auth events which allow us to authenticate the # conflicted state, picking only from the unconflicting state. @@ -515,6 +663,8 @@ def _resolve_with_state_fac(unconflicted_state, conflicted_state, new_needed_events = set(auth_events.itervalues()) new_needed_events -= needed_events + if event_map is not None: + new_needed_events -= set(event_map.iterkeys()) logger.info("Asking for %d auth events", len(new_needed_events)) diff --git a/synapse/storage/__init__.py b/synapse/storage/__init__.py index d01d46338a..de00cae447 100644 --- a/synapse/storage/__init__.py +++ b/synapse/storage/__init__.py @@ -1,5 +1,6 @@ # -*- coding: utf-8 -*- # Copyright 2014-2016 OpenMarket Ltd +# Copyright 2018 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. @@ -19,7 +20,6 @@ from synapse.storage.devices import DeviceStore from .appservice import ( ApplicationServiceStore, ApplicationServiceTransactionStore ) -from ._base import LoggingTransaction from .directory import DirectoryStore from .events import EventsStore from .presence import PresenceStore, UserPresenceState @@ -104,12 +104,6 @@ class DataStore(RoomMemberStore, RoomStore, db_conn, "events", "stream_ordering", step=-1, extra_tables=[("ex_outlier_stream", "event_stream_ordering")] ) - self._receipts_id_gen = StreamIdGenerator( - db_conn, "receipts_linearized", "stream_id" - ) - self._account_data_id_gen = StreamIdGenerator( - db_conn, "account_data_max_stream_id", "stream_id" - ) self._presence_id_gen = StreamIdGenerator( db_conn, "presence_stream", "stream_id" ) @@ -124,7 +118,6 @@ class DataStore(RoomMemberStore, RoomStore, ) self._transaction_id_gen = IdGenerator(db_conn, "sent_transactions", "id") - self._state_groups_id_gen = IdGenerator(db_conn, "state_groups", "id") self._access_tokens_id_gen = IdGenerator(db_conn, "access_tokens", "id") self._event_reports_id_gen = IdGenerator(db_conn, "event_reports", "id") self._push_rule_id_gen = IdGenerator(db_conn, "push_rules", "id") @@ -147,27 +140,6 @@ class DataStore(RoomMemberStore, RoomStore, else: self._cache_id_gen = None - events_max = self._stream_id_gen.get_current_token() - event_cache_prefill, min_event_val = self._get_cache_dict( - db_conn, "events", - entity_column="room_id", - stream_column="stream_ordering", - max_value=events_max, - ) - self._events_stream_cache = StreamChangeCache( - "EventsRoomStreamChangeCache", min_event_val, - prefilled_cache=event_cache_prefill, - ) - - self._membership_stream_cache = StreamChangeCache( - "MembershipStreamChangeCache", events_max, - ) - - account_max = self._account_data_id_gen.get_current_token() - self._account_data_stream_cache = StreamChangeCache( - "AccountDataAndTagsChangeCache", account_max, - ) - self._presence_on_startup = self._get_active_presence(db_conn) presence_cache_prefill, min_presence_val = self._get_cache_dict( @@ -181,18 +153,6 @@ class DataStore(RoomMemberStore, RoomStore, prefilled_cache=presence_cache_prefill ) - push_rules_prefill, push_rules_id = self._get_cache_dict( - db_conn, "push_rules_stream", - entity_column="user_id", - stream_column="stream_id", - max_value=self._push_rules_stream_id_gen.get_current_token()[0], - ) - - self.push_rules_stream_cache = StreamChangeCache( - "PushRulesStreamChangeCache", push_rules_id, - prefilled_cache=push_rules_prefill, - ) - max_device_inbox_id = self._device_inbox_id_gen.get_current_token() device_inbox_prefill, min_device_inbox_id = self._get_cache_dict( db_conn, "device_inbox", @@ -227,6 +187,7 @@ class DataStore(RoomMemberStore, RoomStore, "DeviceListFederationStreamChangeCache", device_list_max, ) + events_max = self._stream_id_gen.get_current_token() curr_state_delta_prefill, min_curr_state_delta_id = self._get_cache_dict( db_conn, "current_state_delta_stream", entity_column="room_id", @@ -251,20 +212,6 @@ class DataStore(RoomMemberStore, RoomStore, prefilled_cache=_group_updates_prefill, ) - cur = LoggingTransaction( - db_conn.cursor(), - name="_find_stream_orderings_for_times_txn", - database_engine=self.database_engine, - after_callbacks=[], - final_callbacks=[], - ) - self._find_stream_orderings_for_times_txn(cur) - cur.close() - - self.find_stream_orderings_looping_call = self._clock.looping_call( - self._find_stream_orderings_for_times, 10 * 60 * 1000 - ) - self._stream_order_on_start = self.get_room_max_stream_ordering() self._min_stream_order_on_start = self.get_room_min_stream_ordering() diff --git a/synapse/storage/_base.py b/synapse/storage/_base.py index b971f0cb18..2fbebd4907 100644 --- a/synapse/storage/_base.py +++ b/synapse/storage/_base.py @@ -48,16 +48,16 @@ class LoggingTransaction(object): passed to the constructor. Adds logging and metrics to the .execute() method.""" __slots__ = [ - "txn", "name", "database_engine", "after_callbacks", "final_callbacks", + "txn", "name", "database_engine", "after_callbacks", "exception_callbacks", ] def __init__(self, txn, name, database_engine, after_callbacks, - final_callbacks): + exception_callbacks): object.__setattr__(self, "txn", txn) object.__setattr__(self, "name", name) object.__setattr__(self, "database_engine", database_engine) object.__setattr__(self, "after_callbacks", after_callbacks) - object.__setattr__(self, "final_callbacks", final_callbacks) + object.__setattr__(self, "exception_callbacks", exception_callbacks) def call_after(self, callback, *args, **kwargs): """Call the given callback on the main twisted thread after the @@ -66,8 +66,8 @@ class LoggingTransaction(object): """ self.after_callbacks.append((callback, args, kwargs)) - def call_finally(self, callback, *args, **kwargs): - self.final_callbacks.append((callback, args, kwargs)) + def call_on_exception(self, callback, *args, **kwargs): + self.exception_callbacks.append((callback, args, kwargs)) def __getattr__(self, name): return getattr(self.txn, name) @@ -215,7 +215,7 @@ class SQLBaseStore(object): self._clock.looping_call(loop, 10000) - def _new_transaction(self, conn, desc, after_callbacks, final_callbacks, + def _new_transaction(self, conn, desc, after_callbacks, exception_callbacks, logging_context, func, *args, **kwargs): start = time.time() * 1000 txn_id = self._TXN_ID @@ -236,7 +236,7 @@ class SQLBaseStore(object): txn = conn.cursor() txn = LoggingTransaction( txn, name, self.database_engine, after_callbacks, - final_callbacks, + exception_callbacks, ) r = func(txn, *args, **kwargs) conn.commit() @@ -291,52 +291,66 @@ class SQLBaseStore(object): @defer.inlineCallbacks def runInteraction(self, desc, func, *args, **kwargs): - """Wraps the .runInteraction() method on the underlying db_pool.""" - current_context = LoggingContext.current_context() + """Starts a transaction on the database and runs a given function - start_time = time.time() * 1000 + Arguments: + desc (str): description of the transaction, for logging and metrics + func (func): callback function, which will be called with a + database transaction (twisted.enterprise.adbapi.Transaction) as + its first argument, followed by `args` and `kwargs`. - after_callbacks = [] - final_callbacks = [] + args (list): positional args to pass to `func` + kwargs (dict): named args to pass to `func` - def inner_func(conn, *args, **kwargs): - with LoggingContext("runInteraction") as context: - sql_scheduling_timer.inc_by(time.time() * 1000 - start_time) + Returns: + Deferred: The result of func + """ + current_context = LoggingContext.current_context() - if self.database_engine.is_connection_closed(conn): - logger.debug("Reconnecting closed database connection") - conn.reconnect() + after_callbacks = [] + exception_callbacks = [] - current_context.copy_to(context) - return self._new_transaction( - conn, desc, after_callbacks, final_callbacks, current_context, - func, *args, **kwargs - ) + def inner_func(conn, *args, **kwargs): + return self._new_transaction( + conn, desc, after_callbacks, exception_callbacks, current_context, + func, *args, **kwargs + ) try: - with PreserveLoggingContext(): - result = yield self._db_pool.runWithConnection( - inner_func, *args, **kwargs - ) + result = yield self.runWithConnection(inner_func, *args, **kwargs) for after_callback, after_args, after_kwargs in after_callbacks: after_callback(*after_args, **after_kwargs) - finally: - for after_callback, after_args, after_kwargs in final_callbacks: + except: # noqa: E722, as we reraise the exception this is fine. + for after_callback, after_args, after_kwargs in exception_callbacks: after_callback(*after_args, **after_kwargs) + raise defer.returnValue(result) @defer.inlineCallbacks def runWithConnection(self, func, *args, **kwargs): - """Wraps the .runInteraction() method on the underlying db_pool.""" + """Wraps the .runWithConnection() method on the underlying db_pool. + + Arguments: + func (func): callback function, which will be called with a + database connection (twisted.enterprise.adbapi.Connection) as + its first argument, followed by `args` and `kwargs`. + args (list): positional args to pass to `func` + kwargs (dict): named args to pass to `func` + + Returns: + Deferred: The result of func + """ current_context = LoggingContext.current_context() start_time = time.time() * 1000 def inner_func(conn, *args, **kwargs): with LoggingContext("runWithConnection") as context: - sql_scheduling_timer.inc_by(time.time() * 1000 - start_time) + sched_duration_ms = time.time() * 1000 - start_time + sql_scheduling_timer.inc_by(sched_duration_ms) + current_context.add_database_scheduled(sched_duration_ms) if self.database_engine.is_connection_closed(conn): logger.debug("Reconnecting closed database connection") @@ -987,7 +1001,8 @@ class SQLBaseStore(object): # __exit__ called after the transaction finishes. ctx = self._cache_id_gen.get_next() stream_id = ctx.__enter__() - txn.call_finally(ctx.__exit__, None, None, None) + txn.call_on_exception(ctx.__exit__, None, None, None) + txn.call_after(ctx.__exit__, None, None, None) txn.call_after(self.hs.get_notifier().on_new_replication_data) self._simple_insert_txn( diff --git a/synapse/storage/account_data.py b/synapse/storage/account_data.py index 40a2ad8d05..f83ff0454a 100644 --- a/synapse/storage/account_data.py +++ b/synapse/storage/account_data.py @@ -1,5 +1,6 @@ # -*- coding: utf-8 -*- # Copyright 2014-2016 OpenMarket Ltd +# Copyright 2018 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. @@ -13,18 +14,46 @@ # See the License for the specific language governing permissions and # limitations under the License. -from ._base import SQLBaseStore from twisted.internet import defer +from synapse.storage._base import SQLBaseStore +from synapse.storage.util.id_generators import StreamIdGenerator + +from synapse.util.caches.stream_change_cache import StreamChangeCache from synapse.util.caches.descriptors import cached, cachedList, cachedInlineCallbacks +import abc import simplejson as json import logging logger = logging.getLogger(__name__) -class AccountDataStore(SQLBaseStore): +class AccountDataWorkerStore(SQLBaseStore): + """This is an abstract base class where subclasses must implement + `get_max_account_data_stream_id` which can be called in the initializer. + """ + + # This ABCMeta metaclass ensures that we cannot be instantiated without + # the abstract methods being implemented. + __metaclass__ = abc.ABCMeta + + def __init__(self, db_conn, hs): + account_max = self.get_max_account_data_stream_id() + self._account_data_stream_cache = StreamChangeCache( + "AccountDataAndTagsChangeCache", account_max, + ) + + super(AccountDataWorkerStore, self).__init__(db_conn, hs) + + @abc.abstractmethod + def get_max_account_data_stream_id(self): + """Get the current max stream ID for account data stream + + Returns: + int + """ + raise NotImplementedError() @cached() def get_account_data_for_user(self, user_id): @@ -104,6 +133,7 @@ class AccountDataStore(SQLBaseStore): for row in rows }) + @cached(num_args=2) def get_account_data_for_room(self, user_id, room_id): """Get all the client account_data for a user for a room. @@ -127,6 +157,38 @@ class AccountDataStore(SQLBaseStore): "get_account_data_for_room", get_account_data_for_room_txn ) + @cached(num_args=3, max_entries=5000) + def get_account_data_for_room_and_type(self, user_id, room_id, account_data_type): + """Get the client account_data of given type for a user for a room. + + Args: + user_id(str): The user to get the account_data for. + room_id(str): The room to get the account_data for. + account_data_type (str): The account data type to get. + Returns: + A deferred of the room account_data for that type, or None if + there isn't any set. + """ + def get_account_data_for_room_and_type_txn(txn): + content_json = self._simple_select_one_onecol_txn( + txn, + table="room_account_data", + keyvalues={ + "user_id": user_id, + "room_id": room_id, + "account_data_type": account_data_type, + }, + retcol="content", + allow_none=True + ) + + return json.loads(content_json) if content_json else None + + return self.runInteraction( + "get_account_data_for_room_and_type", + get_account_data_for_room_and_type_txn, + ) + def get_all_updated_account_data(self, last_global_id, last_room_id, current_id, limit): """Get all the client account_data that has changed on the server @@ -209,6 +271,36 @@ class AccountDataStore(SQLBaseStore): "get_updated_account_data_for_user", get_updated_account_data_for_user_txn ) + @cachedInlineCallbacks(num_args=2, cache_context=True, max_entries=5000) + def is_ignored_by(self, ignored_user_id, ignorer_user_id, cache_context): + ignored_account_data = yield self.get_global_account_data_by_type_for_user( + "m.ignored_user_list", ignorer_user_id, + on_invalidate=cache_context.invalidate, + ) + if not ignored_account_data: + defer.returnValue(False) + + defer.returnValue( + ignored_user_id in ignored_account_data.get("ignored_users", {}) + ) + + +class AccountDataStore(AccountDataWorkerStore): + def __init__(self, db_conn, hs): + self._account_data_id_gen = StreamIdGenerator( + db_conn, "account_data_max_stream_id", "stream_id" + ) + + super(AccountDataStore, self).__init__(db_conn, hs) + + def get_max_account_data_stream_id(self): + """Get the current max stream id for the private user data stream + + Returns: + A deferred int. + """ + return self._account_data_id_gen.get_current_token() + @defer.inlineCallbacks def add_account_data_to_room(self, user_id, room_id, account_data_type, content): """Add some account_data to a room for a user. @@ -251,6 +343,10 @@ class AccountDataStore(SQLBaseStore): self._account_data_stream_cache.entity_has_changed(user_id, next_id) self.get_account_data_for_user.invalidate((user_id,)) + self.get_account_data_for_room.invalidate((user_id, room_id,)) + self.get_account_data_for_room_and_type.prefill( + (user_id, room_id, account_data_type,), content, + ) result = self._account_data_id_gen.get_current_token() defer.returnValue(result) @@ -321,16 +417,3 @@ class AccountDataStore(SQLBaseStore): "update_account_data_max_stream_id", _update, ) - - @cachedInlineCallbacks(num_args=2, cache_context=True, max_entries=5000) - def is_ignored_by(self, ignored_user_id, ignorer_user_id, cache_context): - ignored_account_data = yield self.get_global_account_data_by_type_for_user( - "m.ignored_user_list", ignorer_user_id, - on_invalidate=cache_context.invalidate, - ) - if not ignored_account_data: - defer.returnValue(False) - - defer.returnValue( - ignored_user_id in ignored_account_data.get("ignored_users", {}) - ) diff --git a/synapse/storage/appservice.py b/synapse/storage/appservice.py index d8c84b7141..12ea8a158c 100644 --- a/synapse/storage/appservice.py +++ b/synapse/storage/appservice.py @@ -1,5 +1,6 @@ # -*- coding: utf-8 -*- # Copyright 2015, 2016 OpenMarket Ltd +# Copyright 2018 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. @@ -17,10 +18,9 @@ import re import simplejson as json from twisted.internet import defer -from synapse.api.constants import Membership from synapse.appservice import AppServiceTransaction from synapse.config.appservice import load_appservices -from synapse.storage.roommember import RoomsForUser +from synapse.storage.events import EventsWorkerStore from ._base import SQLBaseStore @@ -46,17 +46,16 @@ def _make_exclusive_regex(services_cache): return exclusive_user_regex -class ApplicationServiceStore(SQLBaseStore): - +class ApplicationServiceWorkerStore(SQLBaseStore): def __init__(self, db_conn, hs): - super(ApplicationServiceStore, self).__init__(db_conn, hs) - self.hostname = hs.hostname self.services_cache = load_appservices( hs.hostname, hs.config.app_service_config_files ) self.exclusive_user_regex = _make_exclusive_regex(self.services_cache) + super(ApplicationServiceWorkerStore, self).__init__(db_conn, hs) + def get_app_services(self): return self.services_cache @@ -99,83 +98,30 @@ class ApplicationServiceStore(SQLBaseStore): return service return None - def get_app_service_rooms(self, service): - """Get a list of RoomsForUser for this application service. - - Application services may be "interested" in lots of rooms depending on - the room ID, the room aliases, or the members in the room. This function - takes all of these into account and returns a list of RoomsForUser which - represent the entire list of room IDs that this application service - wants to know about. + def get_app_service_by_id(self, as_id): + """Get the application service with the given appservice ID. Args: - service: The application service to get a room list for. + as_id (str): The application service ID. Returns: - A list of RoomsForUser. + synapse.appservice.ApplicationService or None. """ - return self.runInteraction( - "get_app_service_rooms", - self._get_app_service_rooms_txn, - service, - ) - - def _get_app_service_rooms_txn(self, txn, service): - # get all rooms matching the room ID regex. - room_entries = self._simple_select_list_txn( - txn=txn, table="rooms", keyvalues=None, retcols=["room_id"] - ) - matching_room_list = set([ - r["room_id"] for r in room_entries if - service.is_interested_in_room(r["room_id"]) - ]) - - # resolve room IDs for matching room alias regex. - room_alias_mappings = self._simple_select_list_txn( - txn=txn, table="room_aliases", keyvalues=None, - retcols=["room_id", "room_alias"] - ) - matching_room_list |= set([ - r["room_id"] for r in room_alias_mappings if - service.is_interested_in_alias(r["room_alias"]) - ]) - - # get all rooms for every user for this AS. This is scoped to users on - # this HS only. - user_list = self._simple_select_list_txn( - txn=txn, table="users", keyvalues=None, retcols=["name"] - ) - user_list = [ - u["name"] for u in user_list if - service.is_interested_in_user(u["name"]) - ] - rooms_for_user_matching_user_id = set() # RoomsForUser list - for user_id in user_list: - # FIXME: This assumes this store is linked with RoomMemberStore :( - rooms_for_user = self._get_rooms_for_user_where_membership_is_txn( - txn=txn, - user_id=user_id, - membership_list=[Membership.JOIN] - ) - rooms_for_user_matching_user_id |= set(rooms_for_user) - - # make RoomsForUser tuples for room ids and aliases which are not in the - # main rooms_for_user_list - e.g. they are rooms which do not have AS - # registered users in it. - known_room_ids = [r.room_id for r in rooms_for_user_matching_user_id] - missing_rooms_for_user = [ - RoomsForUser(r, service.sender, "join") for r in - matching_room_list if r not in known_room_ids - ] - rooms_for_user_matching_user_id |= set(missing_rooms_for_user) - - return rooms_for_user_matching_user_id + for service in self.services_cache: + if service.id == as_id: + return service + return None -class ApplicationServiceTransactionStore(SQLBaseStore): +class ApplicationServiceStore(ApplicationServiceWorkerStore): + # This is currently empty due to there not being any AS storage functions + # that can't be run on the workers. Since this may change in future, and + # to keep consistency with the other stores, we keep this empty class for + # now. + pass - def __init__(self, db_conn, hs): - super(ApplicationServiceTransactionStore, self).__init__(db_conn, hs) +class ApplicationServiceTransactionWorkerStore(ApplicationServiceWorkerStore, + EventsWorkerStore): @defer.inlineCallbacks def get_appservices_by_state(self, state): """Get a list of application services based on their state. @@ -420,3 +366,11 @@ class ApplicationServiceTransactionStore(SQLBaseStore): events = yield self._get_events(event_ids) defer.returnValue((upper_bound, events)) + + +class ApplicationServiceTransactionStore(ApplicationServiceTransactionWorkerStore): + # This is currently empty due to there not being any AS storage functions + # that can't be run on the workers. Since this may change in future, and + # to keep consistency with the other stores, we keep this empty class for + # now. + pass diff --git a/synapse/storage/background_updates.py b/synapse/storage/background_updates.py index 8f3bff311a..8af325a9f5 100644 --- a/synapse/storage/background_updates.py +++ b/synapse/storage/background_updates.py @@ -242,6 +242,25 @@ class BackgroundUpdateStore(SQLBaseStore): """ self._background_update_handlers[update_name] = update_handler + def register_noop_background_update(self, update_name): + """Register a noop handler for a background update. + + This is useful when we previously did a background update, but no + longer wish to do the update. In this case the background update should + be removed from the schema delta files, but there may still be some + users who have the background update queued, so this method should + also be called to clear the update. + + Args: + update_name (str): Name of update + """ + @defer.inlineCallbacks + def noop_update(progress, batch_size): + yield self._end_background_update(update_name) + defer.returnValue(1) + + self.register_background_update_handler(update_name, noop_update) + def register_background_index_update(self, update_name, index_name, table, columns, where_clause=None, unique=False, diff --git a/synapse/storage/directory.py b/synapse/storage/directory.py index 79e7c540ad..d0c0059757 100644 --- a/synapse/storage/directory.py +++ b/synapse/storage/directory.py @@ -29,8 +29,7 @@ RoomAliasMapping = namedtuple( ) -class DirectoryStore(SQLBaseStore): - +class DirectoryWorkerStore(SQLBaseStore): @defer.inlineCallbacks def get_association_from_room_alias(self, room_alias): """ Get's the room_id and server list for a given room_alias @@ -69,6 +68,28 @@ class DirectoryStore(SQLBaseStore): RoomAliasMapping(room_id, room_alias.to_string(), servers) ) + def get_room_alias_creator(self, room_alias): + return self._simple_select_one_onecol( + table="room_aliases", + keyvalues={ + "room_alias": room_alias, + }, + retcol="creator", + desc="get_room_alias_creator", + allow_none=True + ) + + @cached(max_entries=5000) + def get_aliases_for_room(self, room_id): + return self._simple_select_onecol( + "room_aliases", + {"room_id": room_id}, + "room_alias", + desc="get_aliases_for_room", + ) + + +class DirectoryStore(DirectoryWorkerStore): @defer.inlineCallbacks def create_room_alias_association(self, room_alias, room_id, servers, creator=None): """ Creates an associatin between a room alias and room_id/servers @@ -116,17 +137,6 @@ class DirectoryStore(SQLBaseStore): ) defer.returnValue(ret) - def get_room_alias_creator(self, room_alias): - return self._simple_select_one_onecol( - table="room_aliases", - keyvalues={ - "room_alias": room_alias, - }, - retcol="creator", - desc="get_room_alias_creator", - allow_none=True - ) - @defer.inlineCallbacks def delete_room_alias(self, room_alias): room_id = yield self.runInteraction( @@ -135,7 +145,6 @@ class DirectoryStore(SQLBaseStore): room_alias, ) - self.get_aliases_for_room.invalidate((room_id,)) defer.returnValue(room_id) def _delete_room_alias_txn(self, txn, room_alias): @@ -160,17 +169,12 @@ class DirectoryStore(SQLBaseStore): (room_alias.to_string(),) ) - return room_id - - @cached(max_entries=5000) - def get_aliases_for_room(self, room_id): - return self._simple_select_onecol( - "room_aliases", - {"room_id": room_id}, - "room_alias", - desc="get_aliases_for_room", + self._invalidate_cache_and_stream( + txn, self.get_aliases_for_room, (room_id,) ) + return room_id + def update_aliases_for_room(self, old_room_id, new_room_id, creator): def _update_aliases_for_room_txn(txn): sql = "UPDATE room_aliases SET room_id = ?, creator = ? WHERE room_id = ?" diff --git a/synapse/storage/engines/postgres.py b/synapse/storage/engines/postgres.py index a6ae79dfad..8a0386c1a4 100644 --- a/synapse/storage/engines/postgres.py +++ b/synapse/storage/engines/postgres.py @@ -62,3 +62,9 @@ class PostgresEngine(object): def lock_table(self, txn, table): txn.execute("LOCK TABLE %s in EXCLUSIVE MODE" % (table,)) + + def get_next_state_group_id(self, txn): + """Returns an int that can be used as a new state_group ID + """ + txn.execute("SELECT nextval('state_group_id_seq')") + return txn.fetchone()[0] diff --git a/synapse/storage/engines/sqlite3.py b/synapse/storage/engines/sqlite3.py index 755c9a1f07..60f0fa7fb3 100644 --- a/synapse/storage/engines/sqlite3.py +++ b/synapse/storage/engines/sqlite3.py @@ -16,6 +16,7 @@ from synapse.storage.prepare_database import prepare_database import struct +import threading class Sqlite3Engine(object): @@ -24,6 +25,11 @@ class Sqlite3Engine(object): def __init__(self, database_module, database_config): self.module = database_module + # The current max state_group, or None if we haven't looked + # in the DB yet. + self._current_state_group_id = None + self._current_state_group_id_lock = threading.Lock() + def check_database(self, txn): pass @@ -43,6 +49,19 @@ class Sqlite3Engine(object): def lock_table(self, txn, table): return + def get_next_state_group_id(self, txn): + """Returns an int that can be used as a new state_group ID + """ + # We do application locking here since if we're using sqlite then + # we are a single process synapse. + with self._current_state_group_id_lock: + if self._current_state_group_id is None: + txn.execute("SELECT COALESCE(max(id), 0) FROM state_groups") + self._current_state_group_id = txn.fetchone()[0] + + self._current_state_group_id += 1 + return self._current_state_group_id + # Following functions taken from: https://github.com/coleifer/peewee diff --git a/synapse/storage/event_federation.py b/synapse/storage/event_federation.py index 55a05c59d5..00ee82d300 100644 --- a/synapse/storage/event_federation.py +++ b/synapse/storage/event_federation.py @@ -15,7 +15,10 @@ from twisted.internet import defer -from ._base import SQLBaseStore +from synapse.storage._base import SQLBaseStore +from synapse.storage.events import EventsWorkerStore +from synapse.storage.signatures import SignatureWorkerStore + from synapse.api.errors import StoreError from synapse.util.caches.descriptors import cached from unpaddedbase64 import encode_base64 @@ -27,30 +30,8 @@ from Queue import PriorityQueue, Empty logger = logging.getLogger(__name__) -class EventFederationStore(SQLBaseStore): - """ Responsible for storing and serving up the various graphs associated - with an event. Including the main event graph and the auth chains for an - event. - - Also has methods for getting the front (latest) and back (oldest) edges - of the event graphs. These are used to generate the parents for new events - and backfilling from another server respectively. - """ - - EVENT_AUTH_STATE_ONLY = "event_auth_state_only" - - def __init__(self, db_conn, hs): - super(EventFederationStore, self).__init__(db_conn, hs) - - self.register_background_update_handler( - self.EVENT_AUTH_STATE_ONLY, - self._background_delete_non_state_event_auth, - ) - - hs.get_clock().looping_call( - self._delete_old_forward_extrem_cache, 60 * 60 * 1000 - ) - +class EventFederationWorkerStore(EventsWorkerStore, SignatureWorkerStore, + SQLBaseStore): def get_auth_chain(self, event_ids, include_given=False): """Get auth events for given event_ids. The events *must* be state events. @@ -228,88 +209,6 @@ class EventFederationStore(SQLBaseStore): return int(min_depth) if min_depth is not None else None - def _update_min_depth_for_room_txn(self, txn, room_id, depth): - min_depth = self._get_min_depth_interaction(txn, room_id) - - if min_depth and depth >= min_depth: - return - - self._simple_upsert_txn( - txn, - table="room_depth", - keyvalues={ - "room_id": room_id, - }, - values={ - "min_depth": depth, - }, - ) - - def _handle_mult_prev_events(self, txn, events): - """ - For the given event, update the event edges table and forward and - backward extremities tables. - """ - self._simple_insert_many_txn( - txn, - table="event_edges", - values=[ - { - "event_id": ev.event_id, - "prev_event_id": e_id, - "room_id": ev.room_id, - "is_state": False, - } - for ev in events - for e_id, _ in ev.prev_events - ], - ) - - self._update_backward_extremeties(txn, events) - - def _update_backward_extremeties(self, txn, events): - """Updates the event_backward_extremities tables based on the new/updated - events being persisted. - - This is called for new events *and* for events that were outliers, but - are now being persisted as non-outliers. - - Forward extremities are handled when we first start persisting the events. - """ - events_by_room = {} - for ev in events: - events_by_room.setdefault(ev.room_id, []).append(ev) - - query = ( - "INSERT INTO event_backward_extremities (event_id, room_id)" - " SELECT ?, ? WHERE NOT EXISTS (" - " SELECT 1 FROM event_backward_extremities" - " WHERE event_id = ? AND room_id = ?" - " )" - " AND NOT EXISTS (" - " SELECT 1 FROM events WHERE event_id = ? AND room_id = ? " - " AND outlier = ?" - " )" - ) - - txn.executemany(query, [ - (e_id, ev.room_id, e_id, ev.room_id, e_id, ev.room_id, False) - for ev in events for e_id, _ in ev.prev_events - if not ev.internal_metadata.is_outlier() - ]) - - query = ( - "DELETE FROM event_backward_extremities" - " WHERE event_id = ? AND room_id = ?" - ) - txn.executemany( - query, - [ - (ev.event_id, ev.room_id) for ev in events - if not ev.internal_metadata.is_outlier() - ] - ) - def get_forward_extremeties_for_room(self, room_id, stream_ordering): """For a given room_id and stream_ordering, return the forward extremeties of the room at that point in "time". @@ -371,28 +270,6 @@ class EventFederationStore(SQLBaseStore): get_forward_extremeties_for_room_txn ) - def _delete_old_forward_extrem_cache(self): - def _delete_old_forward_extrem_cache_txn(txn): - # Delete entries older than a month, while making sure we don't delete - # the only entries for a room. - sql = (""" - DELETE FROM stream_ordering_to_exterm - WHERE - room_id IN ( - SELECT room_id - FROM stream_ordering_to_exterm - WHERE stream_ordering > ? - ) AND stream_ordering < ? - """) - txn.execute( - sql, - (self.stream_ordering_month_ago, self.stream_ordering_month_ago,) - ) - return self.runInteraction( - "_delete_old_forward_extrem_cache", - _delete_old_forward_extrem_cache_txn - ) - def get_backfill_events(self, room_id, event_list, limit): """Get a list of Events for a given topic that occurred before (and including) the events in event_list. Return a list of max size `limit` @@ -522,6 +399,135 @@ class EventFederationStore(SQLBaseStore): return event_results + +class EventFederationStore(EventFederationWorkerStore): + """ Responsible for storing and serving up the various graphs associated + with an event. Including the main event graph and the auth chains for an + event. + + Also has methods for getting the front (latest) and back (oldest) edges + of the event graphs. These are used to generate the parents for new events + and backfilling from another server respectively. + """ + + EVENT_AUTH_STATE_ONLY = "event_auth_state_only" + + def __init__(self, db_conn, hs): + super(EventFederationStore, self).__init__(db_conn, hs) + + self.register_background_update_handler( + self.EVENT_AUTH_STATE_ONLY, + self._background_delete_non_state_event_auth, + ) + + hs.get_clock().looping_call( + self._delete_old_forward_extrem_cache, 60 * 60 * 1000 + ) + + def _update_min_depth_for_room_txn(self, txn, room_id, depth): + min_depth = self._get_min_depth_interaction(txn, room_id) + + if min_depth and depth >= min_depth: + return + + self._simple_upsert_txn( + txn, + table="room_depth", + keyvalues={ + "room_id": room_id, + }, + values={ + "min_depth": depth, + }, + ) + + def _handle_mult_prev_events(self, txn, events): + """ + For the given event, update the event edges table and forward and + backward extremities tables. + """ + self._simple_insert_many_txn( + txn, + table="event_edges", + values=[ + { + "event_id": ev.event_id, + "prev_event_id": e_id, + "room_id": ev.room_id, + "is_state": False, + } + for ev in events + for e_id, _ in ev.prev_events + ], + ) + + self._update_backward_extremeties(txn, events) + + def _update_backward_extremeties(self, txn, events): + """Updates the event_backward_extremities tables based on the new/updated + events being persisted. + + This is called for new events *and* for events that were outliers, but + are now being persisted as non-outliers. + + Forward extremities are handled when we first start persisting the events. + """ + events_by_room = {} + for ev in events: + events_by_room.setdefault(ev.room_id, []).append(ev) + + query = ( + "INSERT INTO event_backward_extremities (event_id, room_id)" + " SELECT ?, ? WHERE NOT EXISTS (" + " SELECT 1 FROM event_backward_extremities" + " WHERE event_id = ? AND room_id = ?" + " )" + " AND NOT EXISTS (" + " SELECT 1 FROM events WHERE event_id = ? AND room_id = ? " + " AND outlier = ?" + " )" + ) + + txn.executemany(query, [ + (e_id, ev.room_id, e_id, ev.room_id, e_id, ev.room_id, False) + for ev in events for e_id, _ in ev.prev_events + if not ev.internal_metadata.is_outlier() + ]) + + query = ( + "DELETE FROM event_backward_extremities" + " WHERE event_id = ? AND room_id = ?" + ) + txn.executemany( + query, + [ + (ev.event_id, ev.room_id) for ev in events + if not ev.internal_metadata.is_outlier() + ] + ) + + def _delete_old_forward_extrem_cache(self): + def _delete_old_forward_extrem_cache_txn(txn): + # Delete entries older than a month, while making sure we don't delete + # the only entries for a room. + sql = (""" + DELETE FROM stream_ordering_to_exterm + WHERE + room_id IN ( + SELECT room_id + FROM stream_ordering_to_exterm + WHERE stream_ordering > ? + ) AND stream_ordering < ? + """) + txn.execute( + sql, + (self.stream_ordering_month_ago, self.stream_ordering_month_ago,) + ) + return self.runInteraction( + "_delete_old_forward_extrem_cache", + _delete_old_forward_extrem_cache_txn + ) + def clean_room_for_join(self, room_id): return self.runInteraction( "clean_room_for_join", diff --git a/synapse/storage/event_push_actions.py b/synapse/storage/event_push_actions.py index 575d710d5d..e78f8d0114 100644 --- a/synapse/storage/event_push_actions.py +++ b/synapse/storage/event_push_actions.py @@ -1,5 +1,6 @@ # -*- coding: utf-8 -*- # Copyright 2015 OpenMarket Ltd +# Copyright 2018 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. @@ -13,7 +14,7 @@ # See the License for the specific language governing permissions and # limitations under the License. -from ._base import SQLBaseStore +from synapse.storage._base import SQLBaseStore, LoggingTransaction from twisted.internet import defer from synapse.util.async import sleep from synapse.util.caches.descriptors import cachedInlineCallbacks @@ -62,60 +63,28 @@ def _deserialize_action(actions, is_highlight): return DEFAULT_NOTIF_ACTION -class EventPushActionsStore(SQLBaseStore): - EPA_HIGHLIGHT_INDEX = "epa_highlight_index" - +class EventPushActionsWorkerStore(SQLBaseStore): def __init__(self, db_conn, hs): - super(EventPushActionsStore, self).__init__(db_conn, hs) - - self.register_background_index_update( - self.EPA_HIGHLIGHT_INDEX, - index_name="event_push_actions_u_highlight", - table="event_push_actions", - columns=["user_id", "stream_ordering"], + super(EventPushActionsWorkerStore, self).__init__(db_conn, hs) + + # These get correctly set by _find_stream_orderings_for_times_txn + self.stream_ordering_month_ago = None + self.stream_ordering_day_ago = None + + cur = LoggingTransaction( + db_conn.cursor(), + name="_find_stream_orderings_for_times_txn", + database_engine=self.database_engine, + after_callbacks=[], + exception_callbacks=[], ) + self._find_stream_orderings_for_times_txn(cur) + cur.close() - self.register_background_index_update( - "event_push_actions_highlights_index", - index_name="event_push_actions_highlights_index", - table="event_push_actions", - columns=["user_id", "room_id", "topological_ordering", "stream_ordering"], - where_clause="highlight=1" + self.find_stream_orderings_looping_call = self._clock.looping_call( + self._find_stream_orderings_for_times, 10 * 60 * 1000 ) - self._doing_notif_rotation = False - self._rotate_notif_loop = self._clock.looping_call( - self._rotate_notifs, 30 * 60 * 1000 - ) - - def _set_push_actions_for_event_and_users_txn(self, txn, event, tuples): - """ - Args: - event: the event set actions for - tuples: list of tuples of (user_id, actions) - """ - values = [] - for uid, actions in tuples: - is_highlight = 1 if _action_has_highlight(actions) else 0 - - values.append({ - 'room_id': event.room_id, - 'event_id': event.event_id, - 'user_id': uid, - 'actions': _serialize_action(actions, is_highlight), - 'stream_ordering': event.internal_metadata.stream_ordering, - 'topological_ordering': event.depth, - 'notif': 1, - 'highlight': is_highlight, - }) - - for uid, __ in tuples: - txn.call_after( - self.get_unread_event_push_actions_by_room_for_user.invalidate_many, - (event.room_id, uid) - ) - self._simple_insert_many_txn(txn, "event_push_actions", values) - @cachedInlineCallbacks(num_args=3, tree=True, max_entries=5000) def get_unread_event_push_actions_by_room_for_user( self, room_id, user_id, last_read_event_id @@ -432,6 +401,280 @@ class EventPushActionsStore(SQLBaseStore): # Now return the first `limit` defer.returnValue(notifs[:limit]) + def add_push_actions_to_staging(self, event_id, user_id_actions): + """Add the push actions for the event to the push action staging area. + + Args: + event_id (str) + user_id_actions (dict[str, list[dict|str])]): A dictionary mapping + user_id to list of push actions, where an action can either be + a string or dict. + + Returns: + Deferred + """ + + if not user_id_actions: + return + + # This is a helper function for generating the necessary tuple that + # can be used to inert into the `event_push_actions_staging` table. + def _gen_entry(user_id, actions): + is_highlight = 1 if _action_has_highlight(actions) else 0 + return ( + event_id, # event_id column + user_id, # user_id column + _serialize_action(actions, is_highlight), # actions column + 1, # notif column + is_highlight, # highlight column + ) + + def _add_push_actions_to_staging_txn(txn): + # We don't use _simple_insert_many here to avoid the overhead + # of generating lists of dicts. + + sql = """ + INSERT INTO event_push_actions_staging + (event_id, user_id, actions, notif, highlight) + VALUES (?, ?, ?, ?, ?) + """ + + txn.executemany(sql, ( + _gen_entry(user_id, actions) + for user_id, actions in user_id_actions.iteritems() + )) + + return self.runInteraction( + "add_push_actions_to_staging", _add_push_actions_to_staging_txn + ) + + def remove_push_actions_from_staging(self, event_id): + """Called if we failed to persist the event to ensure that stale push + actions don't build up in the DB + + Args: + event_id (str) + """ + + return self._simple_delete( + table="event_push_actions_staging", + keyvalues={ + "event_id": event_id, + }, + desc="remove_push_actions_from_staging", + ) + + @defer.inlineCallbacks + def _find_stream_orderings_for_times(self): + yield self.runInteraction( + "_find_stream_orderings_for_times", + self._find_stream_orderings_for_times_txn + ) + + def _find_stream_orderings_for_times_txn(self, txn): + logger.info("Searching for stream ordering 1 month ago") + self.stream_ordering_month_ago = self._find_first_stream_ordering_after_ts_txn( + txn, self._clock.time_msec() - 30 * 24 * 60 * 60 * 1000 + ) + logger.info( + "Found stream ordering 1 month ago: it's %d", + self.stream_ordering_month_ago + ) + logger.info("Searching for stream ordering 1 day ago") + self.stream_ordering_day_ago = self._find_first_stream_ordering_after_ts_txn( + txn, self._clock.time_msec() - 24 * 60 * 60 * 1000 + ) + logger.info( + "Found stream ordering 1 day ago: it's %d", + self.stream_ordering_day_ago + ) + + def find_first_stream_ordering_after_ts(self, ts): + """Gets the stream ordering corresponding to a given timestamp. + + Specifically, finds the stream_ordering of the first event that was + received on or after the timestamp. This is done by a binary search on + the events table, since there is no index on received_ts, so is + relatively slow. + + Args: + ts (int): timestamp in millis + + Returns: + Deferred[int]: stream ordering of the first event received on/after + the timestamp + """ + return self.runInteraction( + "_find_first_stream_ordering_after_ts_txn", + self._find_first_stream_ordering_after_ts_txn, + ts, + ) + + @staticmethod + def _find_first_stream_ordering_after_ts_txn(txn, ts): + """ + Find the stream_ordering of the first event that was received on or + after a given timestamp. This is relatively slow as there is no index + on received_ts but we can then use this to delete push actions before + this. + + received_ts must necessarily be in the same order as stream_ordering + and stream_ordering is indexed, so we manually binary search using + stream_ordering + + Args: + txn (twisted.enterprise.adbapi.Transaction): + ts (int): timestamp to search for + + Returns: + int: stream ordering + """ + txn.execute("SELECT MAX(stream_ordering) FROM events") + max_stream_ordering = txn.fetchone()[0] + + if max_stream_ordering is None: + return 0 + + # We want the first stream_ordering in which received_ts is greater + # than or equal to ts. Call this point X. + # + # We maintain the invariants: + # + # range_start <= X <= range_end + # + range_start = 0 + range_end = max_stream_ordering + 1 + + # Given a stream_ordering, look up the timestamp at that + # stream_ordering. + # + # The array may be sparse (we may be missing some stream_orderings). + # We treat the gaps as the same as having the same value as the + # preceding entry, because we will pick the lowest stream_ordering + # which satisfies our requirement of received_ts >= ts. + # + # For example, if our array of events indexed by stream_ordering is + # [10, <none>, 20], we should treat this as being equivalent to + # [10, 10, 20]. + # + sql = ( + "SELECT received_ts FROM events" + " WHERE stream_ordering <= ?" + " ORDER BY stream_ordering DESC" + " LIMIT 1" + ) + + while range_end - range_start > 0: + middle = (range_end + range_start) // 2 + txn.execute(sql, (middle,)) + row = txn.fetchone() + if row is None: + # no rows with stream_ordering<=middle + range_start = middle + 1 + continue + + middle_ts = row[0] + if ts > middle_ts: + # we got a timestamp lower than the one we were looking for. + # definitely need to look higher: X > middle. + range_start = middle + 1 + else: + # we got a timestamp higher than (or the same as) the one we + # were looking for. We aren't yet sure about the point we + # looked up, but we can be sure that X <= middle. + range_end = middle + + return range_end + + +class EventPushActionsStore(EventPushActionsWorkerStore): + EPA_HIGHLIGHT_INDEX = "epa_highlight_index" + + def __init__(self, db_conn, hs): + super(EventPushActionsStore, self).__init__(db_conn, hs) + + self.register_background_index_update( + self.EPA_HIGHLIGHT_INDEX, + index_name="event_push_actions_u_highlight", + table="event_push_actions", + columns=["user_id", "stream_ordering"], + ) + + self.register_background_index_update( + "event_push_actions_highlights_index", + index_name="event_push_actions_highlights_index", + table="event_push_actions", + columns=["user_id", "room_id", "topological_ordering", "stream_ordering"], + where_clause="highlight=1" + ) + + self._doing_notif_rotation = False + self._rotate_notif_loop = self._clock.looping_call( + self._rotate_notifs, 30 * 60 * 1000 + ) + + def _set_push_actions_for_event_and_users_txn(self, txn, events_and_contexts, + all_events_and_contexts): + """Handles moving push actions from staging table to main + event_push_actions table for all events in `events_and_contexts`. + + Also ensures that all events in `all_events_and_contexts` are removed + from the push action staging area. + + Args: + events_and_contexts (list[(EventBase, EventContext)]): events + we are persisting + all_events_and_contexts (list[(EventBase, EventContext)]): all + events that we were going to persist. This includes events + we've already persisted, etc, that wouldn't appear in + events_and_context. + """ + + sql = """ + INSERT INTO event_push_actions ( + room_id, event_id, user_id, actions, stream_ordering, + topological_ordering, notif, highlight + ) + SELECT ?, event_id, user_id, actions, ?, ?, notif, highlight + FROM event_push_actions_staging + WHERE event_id = ? + """ + + if events_and_contexts: + txn.executemany(sql, ( + ( + event.room_id, event.internal_metadata.stream_ordering, + event.depth, event.event_id, + ) + for event, _ in events_and_contexts + )) + + for event, _ in events_and_contexts: + user_ids = self._simple_select_onecol_txn( + txn, + table="event_push_actions_staging", + keyvalues={ + "event_id": event.event_id, + }, + retcol="user_id", + ) + + for uid in user_ids: + txn.call_after( + self.get_unread_event_push_actions_by_room_for_user.invalidate_many, + (event.room_id, uid,) + ) + + # Now we delete the staging area for *all* events that were being + # persisted. + txn.executemany( + "DELETE FROM event_push_actions_staging WHERE event_id = ?", + ( + (event.event_id,) + for event, _ in all_events_and_contexts + ) + ) + @defer.inlineCallbacks def get_push_actions_for_user(self, user_id, before=None, limit=50, only_highlight=False): @@ -551,69 +794,6 @@ class EventPushActionsStore(SQLBaseStore): """, (room_id, user_id, stream_ordering)) @defer.inlineCallbacks - def _find_stream_orderings_for_times(self): - yield self.runInteraction( - "_find_stream_orderings_for_times", - self._find_stream_orderings_for_times_txn - ) - - def _find_stream_orderings_for_times_txn(self, txn): - logger.info("Searching for stream ordering 1 month ago") - self.stream_ordering_month_ago = self._find_first_stream_ordering_after_ts_txn( - txn, self._clock.time_msec() - 30 * 24 * 60 * 60 * 1000 - ) - logger.info( - "Found stream ordering 1 month ago: it's %d", - self.stream_ordering_month_ago - ) - logger.info("Searching for stream ordering 1 day ago") - self.stream_ordering_day_ago = self._find_first_stream_ordering_after_ts_txn( - txn, self._clock.time_msec() - 24 * 60 * 60 * 1000 - ) - logger.info( - "Found stream ordering 1 day ago: it's %d", - self.stream_ordering_day_ago - ) - - def _find_first_stream_ordering_after_ts_txn(self, txn, ts): - """ - Find the stream_ordering of the first event that was received after - a given timestamp. This is relatively slow as there is no index on - received_ts but we can then use this to delete push actions before - this. - - received_ts must necessarily be in the same order as stream_ordering - and stream_ordering is indexed, so we manually binary search using - stream_ordering - """ - txn.execute("SELECT MAX(stream_ordering) FROM events") - max_stream_ordering = txn.fetchone()[0] - - if max_stream_ordering is None: - return 0 - - range_start = 0 - range_end = max_stream_ordering - - sql = ( - "SELECT received_ts FROM events" - " WHERE stream_ordering > ?" - " ORDER BY stream_ordering" - " LIMIT 1" - ) - - while range_end - range_start > 1: - middle = int((range_end + range_start) / 2) - txn.execute(sql, (middle,)) - middle_ts = txn.fetchone()[0] - if ts > middle_ts: - range_start = middle - else: - range_end = middle - - return range_end - - @defer.inlineCallbacks def _rotate_notifs(self): if self._doing_notif_rotation or self.stream_ordering_day_ago is None: return diff --git a/synapse/storage/events.py b/synapse/storage/events.py index 73658a9927..85ce6bea1a 100644 --- a/synapse/storage/events.py +++ b/synapse/storage/events.py @@ -1,5 +1,6 @@ # -*- coding: utf-8 -*- # Copyright 2014-2016 OpenMarket Ltd +# Copyright 2018 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. @@ -12,23 +13,22 @@ # 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 SQLBaseStore -from twisted.internet import defer, reactor +from synapse.storage.events_worker import EventsWorkerStore -from synapse.events import FrozenEvent, USE_FROZEN_DICTS -from synapse.events.utils import prune_event +from twisted.internet import defer + +from synapse.events import USE_FROZEN_DICTS from synapse.util.async import ObservableDeferred from synapse.util.logcontext import ( - preserve_fn, PreserveLoggingContext, make_deferred_yieldable + PreserveLoggingContext, make_deferred_yieldable ) from synapse.util.logutils import log_function from synapse.util.metrics import Measure from synapse.api.constants import EventTypes from synapse.api.errors import SynapseError -from synapse.state import resolve_events -from synapse.util.caches.descriptors import cached +from synapse.util.caches.descriptors import cached, cachedInlineCallbacks from synapse.types import get_domain_from_id from canonicaljson import encode_canonical_json @@ -61,16 +61,6 @@ def encode_json(json_object): return json.dumps(json_object, ensure_ascii=False) -# These values are used in the `enqueus_event` and `_do_fetch` methods to -# control how we batch/bulk fetch events from the database. -# The values are plucked out of thing air to make initial sync run faster -# on jki.re -# TODO: Make these configurable. -EVENT_QUEUE_THREADS = 3 # Max number of threads that will fetch events -EVENT_QUEUE_ITERATIONS = 3 # No. times we block waiting for requests for events -EVENT_QUEUE_TIMEOUT_S = 0.1 # Timeout when waiting for requests for events - - class _EventPeristenceQueue(object): """Queues up events so that they can be persisted in bulk with only one concurrent transaction per room. @@ -109,7 +99,7 @@ class _EventPeristenceQueue(object): end_item.events_and_contexts.extend(events_and_contexts) return end_item.deferred.observe() - deferred = ObservableDeferred(defer.Deferred()) + deferred = ObservableDeferred(defer.Deferred(), consumeErrors=True) queue.append(self._EventPersistQueueItem( events_and_contexts=events_and_contexts, @@ -145,18 +135,25 @@ class _EventPeristenceQueue(object): try: queue = self._get_drainining_queue(room_id) for item in queue: + # handle_queue_loop runs in the sentinel logcontext, so + # there is no need to preserve_fn when running the + # callbacks on the deferred. try: ret = yield per_item_callback(item) item.deferred.callback(ret) - except Exception as e: - item.deferred.errback(e) + except Exception: + item.deferred.errback() finally: queue = self._event_persist_queues.pop(room_id, None) if queue: self._event_persist_queues[room_id] = queue self._currently_persisting_rooms.discard(room_id) - preserve_fn(handle_queue_loop)() + # set handle_queue_loop off on the background. We don't want to + # attribute work done in it to the current request, so we drop the + # logcontext altogether. + with PreserveLoggingContext(): + handle_queue_loop() def _get_drainining_queue(self, room_id): queue = self._event_persist_queues.setdefault(room_id, deque()) @@ -192,13 +189,12 @@ def _retry_on_integrity_error(func): return f -class EventsStore(SQLBaseStore): +class EventsStore(EventsWorkerStore): EVENT_ORIGIN_SERVER_TS_NAME = "event_origin_server_ts" EVENT_FIELDS_SENDER_URL_UPDATE_NAME = "event_fields_sender_url" def __init__(self, db_conn, hs): super(EventsStore, self).__init__(db_conn, hs) - self._clock = hs.get_clock() self.register_background_update_handler( self.EVENT_ORIGIN_SERVER_TS_NAME, self._background_reindex_origin_server_ts ) @@ -229,6 +225,8 @@ class EventsStore(SQLBaseStore): self._event_persist_queue = _EventPeristenceQueue() + self._state_resolution_handler = hs.get_state_resolution_handler() + def persist_events(self, events_and_contexts, backfilled=False): """ Write events to the database @@ -284,10 +282,11 @@ class EventsStore(SQLBaseStore): def _maybe_start_persisting(self, room_id): @defer.inlineCallbacks def persisting_queue(item): - yield self._persist_events( - item.events_and_contexts, - backfilled=item.backfilled, - ) + with Measure(self._clock, "persist_events"): + yield self._persist_events( + item.events_and_contexts, + backfilled=item.backfilled, + ) self._event_persist_queue.handle_queue(room_id, persisting_queue) @@ -334,8 +333,20 @@ class EventsStore(SQLBaseStore): # NB: Assumes that we are only persisting events for one room # at a time. + + # map room_id->list[event_ids] giving the new forward + # extremities in each room new_forward_extremeties = {} + + # map room_id->(type,state_key)->event_id tracking the full + # state in each room after adding these events current_state_for_room = {} + + # map room_id->(to_delete, to_insert) where each entry is + # a map (type,key)->event_id giving the state delta in each + # room + state_delta_for_room = {} + if not backfilled: with Measure(self._clock, "_calculate_state_and_extrem"): # Work out the new "current state" for each room. @@ -378,11 +389,20 @@ class EventsStore(SQLBaseStore): if all_single_prev_not_state: continue - state = yield self._calculate_state_delta( - room_id, ev_ctx_rm, new_latest_event_ids + logger.info( + "Calculating state delta for room %s", room_id, ) - if state: - current_state_for_room[room_id] = state + current_state = yield self._get_new_state_after_events( + room_id, + ev_ctx_rm, new_latest_event_ids, + ) + if current_state is not None: + current_state_for_room[room_id] = current_state + delta = yield self._calculate_state_delta( + room_id, current_state, + ) + if delta is not None: + state_delta_for_room[room_id] = delta yield self.runInteraction( "persist_events", @@ -390,7 +410,7 @@ class EventsStore(SQLBaseStore): events_and_contexts=chunk, backfilled=backfilled, delete_existing=delete_existing, - current_state_for_room=current_state_for_room, + state_delta_for_room=state_delta_for_room, new_forward_extremeties=new_forward_extremeties, ) persist_event_counter.inc_by(len(chunk)) @@ -407,7 +427,7 @@ class EventsStore(SQLBaseStore): event_counter.inc(event.type, origin_type, origin_entity) - for room_id, (_, _, new_state) in current_state_for_room.iteritems(): + for room_id, new_state in current_state_for_room.iteritems(): self.get_current_state_ids.prefill( (room_id, ), new_state ) @@ -459,22 +479,31 @@ class EventsStore(SQLBaseStore): defer.returnValue(new_latest_event_ids) @defer.inlineCallbacks - def _calculate_state_delta(self, room_id, events_context, new_latest_event_ids): - """Calculate the new state deltas for a room. + def _get_new_state_after_events(self, room_id, events_context, new_latest_event_ids): + """Calculate the current state dict after adding some new events to + a room - Assumes that we are only persisting events for one room at a time. + Args: + room_id (str): + room to which the events are being added. Used for logging etc + + events_context (list[(EventBase, EventContext)]): + events and contexts which are being added to the room + + new_latest_event_ids (iterable[str]): + the new forward extremities for the room. Returns: - 3-tuple (to_delete, to_insert, new_state) where both are state dicts, - i.e. (type, state_key) -> event_id. `to_delete` are the entries to - first be deleted from current_state_events, `to_insert` are entries - to insert. `new_state` is the full set of state. - May return None if there are no changes to be applied. + Deferred[dict[(str,str), str]|None]: + None if there are no changes to the room state, or + a dict of (type, state_key) -> event_id]. """ - # Now we need to work out the different state sets for - # each state extremities - state_sets = [] - state_groups = set() + + if not new_latest_event_ids: + defer.returnValue({}) + + # map from state_group to ((type, key) -> event_id) state map + state_groups = {} missing_event_ids = [] was_updated = False for event_id in new_latest_event_ids: @@ -485,16 +514,19 @@ class EventsStore(SQLBaseStore): if ctx.current_state_ids is None: raise Exception("Unknown current state") + if ctx.state_group is None: + # I don't think this can happen, but let's double-check + raise Exception( + "Context for new extremity event %s has no state " + "group" % (event_id, ), + ) + # If we've already seen the state group don't bother adding # it to the state sets again if ctx.state_group not in state_groups: - state_sets.append(ctx.current_state_ids) + state_groups[ctx.state_group] = ctx.current_state_ids if ctx.delta_ids or hasattr(ev, "state_key"): was_updated = True - if ctx.state_group: - # Add this as a seen state group (if it has a state - # group) - state_groups.add(ctx.state_group) break else: # If we couldn't find it, then we'll need to pull @@ -502,60 +534,50 @@ class EventsStore(SQLBaseStore): was_updated = True missing_event_ids.append(event_id) + if not was_updated: + return + if missing_event_ids: # Now pull out the state for any missing events from DB event_to_groups = yield self._get_state_group_for_events( missing_event_ids, ) - groups = set(event_to_groups.itervalues()) - state_groups + groups = set(event_to_groups.itervalues()) - set(state_groups.iterkeys()) if groups: group_to_state = yield self._get_state_for_groups(groups) - state_sets.extend(group_to_state.itervalues()) + state_groups.update(group_to_state) - if not new_latest_event_ids: - current_state = {} - elif was_updated: - if len(state_sets) == 1: - # If there is only one state set, then we know what the current - # state is. - current_state = state_sets[0] - else: - # We work out the current state by passing the state sets to the - # state resolution algorithm. It may ask for some events, including - # the events we have yet to persist, so we need a slightly more - # complicated event lookup function than simply looking the events - # up in the db. - events_map = {ev.event_id: ev for ev, _ in events_context} - - @defer.inlineCallbacks - def get_events(ev_ids): - # We get the events by first looking at the list of events we - # are trying to persist, and then fetching the rest from the DB. - db = [] - to_return = {} - for ev_id in ev_ids: - ev = events_map.get(ev_id, None) - if ev: - to_return[ev_id] = ev - else: - db.append(ev_id) - - if db: - evs = yield self.get_events( - ev_ids, get_prev_content=False, check_redacted=False, - ) - to_return.update(evs) - defer.returnValue(to_return) + if len(state_groups) == 1: + # If there is only one state group, then we know what the current + # state is. + defer.returnValue(state_groups.values()[0]) - current_state = yield resolve_events( - state_sets, - state_map_factory=get_events, - ) - else: - return + def get_events(ev_ids): + return self.get_events( + ev_ids, get_prev_content=False, check_redacted=False, + ) + events_map = {ev.event_id: ev for ev, _ in events_context} + logger.debug("calling resolve_state_groups from preserve_events") + res = yield self._state_resolution_handler.resolve_state_groups( + room_id, state_groups, events_map, get_events + ) + + defer.returnValue(res.state) + + @defer.inlineCallbacks + def _calculate_state_delta(self, room_id, current_state): + """Calculate the new state deltas for a room. + Assumes that we are only persisting events for one room at a time. + + Returns: + 2-tuple (to_delete, to_insert) where both are state dicts, + i.e. (type, state_key) -> event_id. `to_delete` are the entries to + first be deleted from current_state_events, `to_insert` are entries + to insert. + """ existing_state = yield self.get_current_state_ids(room_id) existing_events = set(existing_state.itervalues()) @@ -575,67 +597,11 @@ class EventsStore(SQLBaseStore): if ev_id in events_to_insert } - defer.returnValue((to_delete, to_insert, current_state)) - - @defer.inlineCallbacks - def get_event(self, event_id, check_redacted=True, - get_prev_content=False, allow_rejected=False, - allow_none=False): - """Get an event from the database by event_id. - - Args: - event_id (str): The event_id of the event to fetch - check_redacted (bool): If True, check if event has been redacted - and redact it. - get_prev_content (bool): If True and event is a state event, - include the previous states content in the unsigned field. - allow_rejected (bool): If True return rejected events. - allow_none (bool): If True, return None if no event found, if - False throw an exception. - - Returns: - Deferred : A FrozenEvent. - """ - events = yield self._get_events( - [event_id], - check_redacted=check_redacted, - get_prev_content=get_prev_content, - allow_rejected=allow_rejected, - ) - - if not events and not allow_none: - raise SynapseError(404, "Could not find event %s" % (event_id,)) - - defer.returnValue(events[0] if events else None) - - @defer.inlineCallbacks - def get_events(self, event_ids, check_redacted=True, - get_prev_content=False, allow_rejected=False): - """Get events from the database - - Args: - event_ids (list): The event_ids of the events to fetch - check_redacted (bool): If True, check if event has been redacted - and redact it. - get_prev_content (bool): If True and event is a state event, - include the previous states content in the unsigned field. - allow_rejected (bool): If True return rejected events. - - Returns: - Deferred : Dict from event_id to event. - """ - events = yield self._get_events( - event_ids, - check_redacted=check_redacted, - get_prev_content=get_prev_content, - allow_rejected=allow_rejected, - ) - - defer.returnValue({e.event_id: e for e in events}) + defer.returnValue((to_delete, to_insert)) @log_function def _persist_events_txn(self, txn, events_and_contexts, backfilled, - delete_existing=False, current_state_for_room={}, + delete_existing=False, state_delta_for_room={}, new_forward_extremeties={}): """Insert some number of room events into the necessary database tables. @@ -651,7 +617,7 @@ class EventsStore(SQLBaseStore): delete_existing (bool): True to purge existing table rows for the events from the database. This is useful when retrying due to IntegrityError. - current_state_for_room (dict[str, (list[str], list[str])]): + state_delta_for_room (dict[str, (list[str], list[str])]): The current-state delta for each room. For each room, a tuple (to_delete, to_insert), being a list of event ids to be removed from the current state, and a list of event ids to be added to @@ -661,9 +627,11 @@ class EventsStore(SQLBaseStore): list of the event ids which are the forward extremities. """ + all_events_and_contexts = events_and_contexts + max_stream_order = events_and_contexts[-1][0].internal_metadata.stream_ordering - self._update_current_state_txn(txn, current_state_for_room, max_stream_order) + self._update_current_state_txn(txn, state_delta_for_room, max_stream_order) self._update_forward_extremities_txn( txn, @@ -707,9 +675,8 @@ class EventsStore(SQLBaseStore): events_and_contexts=events_and_contexts, ) - # Insert into the state_groups, state_groups_state, and - # event_to_state_groups tables. - self._store_mult_state_groups_txn(txn, events_and_contexts) + # Insert into event_to_state_groups. + self._store_event_state_mappings_txn(txn, events_and_contexts) # _store_rejected_events_txn filters out any events which were # rejected, and returns the filtered list. @@ -724,12 +691,13 @@ class EventsStore(SQLBaseStore): self._update_metadata_tables_txn( txn, events_and_contexts=events_and_contexts, + all_events_and_contexts=all_events_and_contexts, backfilled=backfilled, ) def _update_current_state_txn(self, txn, state_delta_by_room, max_stream_order): for room_id, current_state_tuple in state_delta_by_room.iteritems(): - to_delete, to_insert, _ = current_state_tuple + to_delete, to_insert = current_state_tuple txn.executemany( "DELETE FROM current_state_events WHERE event_id = ?", [(ev_id,) for ev_id in to_delete.itervalues()], @@ -786,7 +754,7 @@ class EventsStore(SQLBaseStore): for member in members_changed: self._invalidate_cache_and_stream( - txn, self.get_rooms_for_user, (member,) + txn, self.get_rooms_for_user_with_stream_ordering, (member,) ) for host in set(get_domain_from_id(u) for u in members_changed): @@ -944,10 +912,9 @@ class EventsStore(SQLBaseStore): # an outlier in the database. We now have some state at that # so we need to update the state_groups table with that state. - # insert into the state_group, state_groups_state and - # event_to_state_groups tables. + # insert into event_to_state_groups. try: - self._store_mult_state_groups_txn(txn, ((event, context),)) + self._store_event_state_mappings_txn(txn, ((event, context),)) except Exception: logger.exception("") raise @@ -1122,27 +1089,33 @@ class EventsStore(SQLBaseStore): ec for ec in events_and_contexts if ec[0] not in to_remove ] - def _update_metadata_tables_txn(self, txn, events_and_contexts, backfilled): + def _update_metadata_tables_txn(self, txn, events_and_contexts, + all_events_and_contexts, backfilled): """Update all the miscellaneous tables for new events Args: txn (twisted.enterprise.adbapi.Connection): db connection events_and_contexts (list[(EventBase, EventContext)]): events we are persisting + all_events_and_contexts (list[(EventBase, EventContext)]): all + events that we were going to persist. This includes events + we've already persisted, etc, that wouldn't appear in + events_and_context. backfilled (bool): True if the events were backfilled """ + # Insert all the push actions into the event_push_actions table. + self._set_push_actions_for_event_and_users_txn( + txn, + events_and_contexts=events_and_contexts, + all_events_and_contexts=all_events_and_contexts, + ) + if not events_and_contexts: # nothing to do here return for event, context in events_and_contexts: - # Insert all the push actions into the event_push_actions table. - if context.push_actions: - self._set_push_actions_for_event_and_users_txn( - txn, event, context.push_actions - ) - if event.type == EventTypes.Redaction and event.redacts is not None: # Remove the entries in the event_push_actions table for the # redacted event. @@ -1347,292 +1320,6 @@ class EventsStore(SQLBaseStore): ) @defer.inlineCallbacks - def _get_events(self, event_ids, check_redacted=True, - get_prev_content=False, allow_rejected=False): - if not event_ids: - defer.returnValue([]) - - event_id_list = event_ids - event_ids = set(event_ids) - - event_entry_map = self._get_events_from_cache( - event_ids, - allow_rejected=allow_rejected, - ) - - missing_events_ids = [e for e in event_ids if e not in event_entry_map] - - if missing_events_ids: - missing_events = yield self._enqueue_events( - missing_events_ids, - check_redacted=check_redacted, - allow_rejected=allow_rejected, - ) - - event_entry_map.update(missing_events) - - events = [] - for event_id in event_id_list: - entry = event_entry_map.get(event_id, None) - if not entry: - continue - - if allow_rejected or not entry.event.rejected_reason: - if check_redacted and entry.redacted_event: - event = entry.redacted_event - else: - event = entry.event - - events.append(event) - - if get_prev_content: - if "replaces_state" in event.unsigned: - prev = yield self.get_event( - event.unsigned["replaces_state"], - get_prev_content=False, - allow_none=True, - ) - if prev: - event.unsigned = dict(event.unsigned) - event.unsigned["prev_content"] = prev.content - event.unsigned["prev_sender"] = prev.sender - - defer.returnValue(events) - - def _invalidate_get_event_cache(self, event_id): - self._get_event_cache.invalidate((event_id,)) - - def _get_events_from_cache(self, events, allow_rejected, update_metrics=True): - """Fetch events from the caches - - Args: - events (list(str)): list of event_ids to fetch - allow_rejected (bool): Whether to teturn events that were rejected - update_metrics (bool): Whether to update the cache hit ratio metrics - - Returns: - dict of event_id -> _EventCacheEntry for each event_id in cache. If - allow_rejected is `False` then there will still be an entry but it - will be `None` - """ - event_map = {} - - for event_id in events: - ret = self._get_event_cache.get( - (event_id,), None, - update_metrics=update_metrics, - ) - if not ret: - continue - - if allow_rejected or not ret.event.rejected_reason: - event_map[event_id] = ret - else: - event_map[event_id] = None - - return event_map - - def _do_fetch(self, conn): - """Takes a database connection and waits for requests for events from - the _event_fetch_list queue. - """ - event_list = [] - i = 0 - while True: - try: - with self._event_fetch_lock: - event_list = self._event_fetch_list - self._event_fetch_list = [] - - if not event_list: - single_threaded = self.database_engine.single_threaded - if single_threaded or i > EVENT_QUEUE_ITERATIONS: - self._event_fetch_ongoing -= 1 - return - else: - self._event_fetch_lock.wait(EVENT_QUEUE_TIMEOUT_S) - i += 1 - continue - i = 0 - - event_id_lists = zip(*event_list)[0] - event_ids = [ - item for sublist in event_id_lists for item in sublist - ] - - rows = self._new_transaction( - conn, "do_fetch", [], [], None, self._fetch_event_rows, event_ids - ) - - row_dict = { - r["event_id"]: r - for r in rows - } - - # We only want to resolve deferreds from the main thread - def fire(lst, res): - for ids, d in lst: - if not d.called: - try: - with PreserveLoggingContext(): - d.callback([ - res[i] - for i in ids - if i in res - ]) - except Exception: - logger.exception("Failed to callback") - with PreserveLoggingContext(): - reactor.callFromThread(fire, event_list, row_dict) - except Exception as e: - logger.exception("do_fetch") - - # We only want to resolve deferreds from the main thread - def fire(evs): - for _, d in evs: - if not d.called: - with PreserveLoggingContext(): - d.errback(e) - - if event_list: - with PreserveLoggingContext(): - reactor.callFromThread(fire, event_list) - - @defer.inlineCallbacks - def _enqueue_events(self, events, check_redacted=True, allow_rejected=False): - """Fetches events from the database using the _event_fetch_list. This - allows batch and bulk fetching of events - it allows us to fetch events - without having to create a new transaction for each request for events. - """ - if not events: - defer.returnValue({}) - - events_d = defer.Deferred() - with self._event_fetch_lock: - self._event_fetch_list.append( - (events, events_d) - ) - - self._event_fetch_lock.notify() - - if self._event_fetch_ongoing < EVENT_QUEUE_THREADS: - self._event_fetch_ongoing += 1 - should_start = True - else: - should_start = False - - if should_start: - with PreserveLoggingContext(): - self.runWithConnection( - self._do_fetch - ) - - logger.debug("Loading %d events", len(events)) - with PreserveLoggingContext(): - rows = yield events_d - logger.debug("Loaded %d events (%d rows)", len(events), len(rows)) - - if not allow_rejected: - rows[:] = [r for r in rows if not r["rejects"]] - - res = yield make_deferred_yieldable(defer.gatherResults( - [ - preserve_fn(self._get_event_from_row)( - row["internal_metadata"], row["json"], row["redacts"], - rejected_reason=row["rejects"], - ) - for row in rows - ], - consumeErrors=True - )) - - defer.returnValue({ - e.event.event_id: e - for e in res if e - }) - - def _fetch_event_rows(self, txn, events): - rows = [] - N = 200 - for i in range(1 + len(events) / N): - evs = events[i * N:(i + 1) * N] - if not evs: - break - - sql = ( - "SELECT " - " e.event_id as event_id, " - " e.internal_metadata," - " e.json," - " r.redacts as redacts," - " rej.event_id as rejects " - " FROM event_json as e" - " LEFT JOIN rejections as rej USING (event_id)" - " LEFT JOIN redactions as r ON e.event_id = r.redacts" - " WHERE e.event_id IN (%s)" - ) % (",".join(["?"] * len(evs)),) - - txn.execute(sql, evs) - rows.extend(self.cursor_to_dict(txn)) - - return rows - - @defer.inlineCallbacks - def _get_event_from_row(self, internal_metadata, js, redacted, - rejected_reason=None): - with Measure(self._clock, "_get_event_from_row"): - d = json.loads(js) - internal_metadata = json.loads(internal_metadata) - - if rejected_reason: - rejected_reason = yield self._simple_select_one_onecol( - table="rejections", - keyvalues={"event_id": rejected_reason}, - retcol="reason", - desc="_get_event_from_row_rejected_reason", - ) - - original_ev = FrozenEvent( - d, - internal_metadata_dict=internal_metadata, - rejected_reason=rejected_reason, - ) - - redacted_event = None - if redacted: - redacted_event = prune_event(original_ev) - - redaction_id = yield self._simple_select_one_onecol( - table="redactions", - keyvalues={"redacts": redacted_event.event_id}, - retcol="event_id", - desc="_get_event_from_row_redactions", - ) - - redacted_event.unsigned["redacted_by"] = redaction_id - # Get the redaction event. - - because = yield self.get_event( - redaction_id, - check_redacted=False, - allow_none=True, - ) - - if because: - # It's fine to do add the event directly, since get_pdu_json - # will serialise this field correctly - redacted_event.unsigned["redacted_because"] = because - - cache_entry = _EventCacheEntry( - event=original_ev, - redacted_event=redacted_event, - ) - - self._get_event_cache.prefill((original_ev.event_id,), cache_entry) - - defer.returnValue(cache_entry) - - @defer.inlineCallbacks def count_daily_messages(self): """ Returns an estimate of the number of messages sent in the last day. @@ -2017,16 +1704,32 @@ class EventsStore(SQLBaseStore): ) return self.runInteraction("get_all_new_events", get_all_new_events_txn) - def delete_old_state(self, room_id, topological_ordering): - return self.runInteraction( - "delete_old_state", - self._delete_old_state_txn, room_id, topological_ordering - ) + def purge_history( + self, room_id, topological_ordering, delete_local_events, + ): + """Deletes room history before a certain point + + Args: + room_id (str): + + topological_ordering (int): + minimum topo ordering to preserve - def _delete_old_state_txn(self, txn, room_id, topological_ordering): - """Deletes old room state + 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). """ + return self.runInteraction( + "purge_history", + self._purge_history_txn, room_id, topological_ordering, + delete_local_events, + ) + + def _purge_history_txn( + self, txn, room_id, topological_ordering, delete_local_events, + ): # Tables that should be pruned: # event_auth # event_backward_extremities @@ -2047,6 +1750,30 @@ class EventsStore(SQLBaseStore): # state_groups # state_groups_state + # we will build a temporary table listing the events so that we don't + # have to keep shovelling the list back and forth across the + # connection. Annoyingly the python sqlite driver commits the + # transaction on CREATE, so let's do this first. + # + # furthermore, we might already have the table from a previous (failed) + # purge attempt, so let's drop the table first. + + txn.execute("DROP TABLE IF EXISTS events_to_purge") + + txn.execute( + "CREATE TEMPORARY TABLE events_to_purge (" + " event_id TEXT NOT NULL," + " should_delete BOOLEAN NOT NULL" + ")" + ) + + # create an index on should_delete because later we'll be looking for + # the should_delete / shouldn't_delete subsets + txn.execute( + "CREATE INDEX events_to_purge_should_delete" + " ON events_to_purge(should_delete)", + ) + # First ensure that we're not about to delete all the forward extremeties txn.execute( "SELECT e.event_id, e.depth FROM events as e " @@ -2067,42 +1794,48 @@ class EventsStore(SQLBaseStore): 400, "topological_ordering is greater than forward extremeties" ) - logger.debug("[purge] looking for events to delete") + logger.info("[purge] looking for events to delete") + + should_delete_expr = "state_key IS NULL" + should_delete_params = () + if not delete_local_events: + should_delete_expr += " AND event_id NOT LIKE ?" + should_delete_params += ("%:" + self.hs.hostname, ) + + should_delete_params += (room_id, topological_ordering) txn.execute( - "SELECT event_id, state_key FROM events" - " LEFT JOIN state_events USING (room_id, event_id)" - " WHERE room_id = ? AND topological_ordering < ?", - (room_id, topological_ordering,) + "INSERT INTO events_to_purge" + " SELECT event_id, %s" + " FROM events AS e LEFT JOIN state_events USING (event_id)" + " WHERE e.room_id = ? AND topological_ordering < ?" % ( + should_delete_expr, + ), + should_delete_params, + ) + txn.execute( + "SELECT event_id, should_delete FROM events_to_purge" ) event_rows = txn.fetchall() - - to_delete = [ - (event_id,) for event_id, state_key in event_rows - if state_key is None and not self.hs.is_mine_id(event_id) - ] logger.info( - "[purge] found %i events before cutoff, of which %i are remote" - " non-state events to delete", len(event_rows), len(to_delete)) - - for event_id, state_key in event_rows: - txn.call_after(self._get_state_group_for_event.invalidate, (event_id,)) + "[purge] found %i events before cutoff, of which %i can be deleted", + len(event_rows), sum(1 for e in event_rows if e[1]), + ) - logger.debug("[purge] Finding new backward extremities") + logger.info("[purge] Finding new backward extremities") # We calculate the new entries for the backward extremeties by finding # all events that point to events that are to be purged txn.execute( - "SELECT DISTINCT e.event_id FROM events as e" - " INNER JOIN event_edges as ed ON e.event_id = ed.prev_event_id" - " INNER JOIN events as e2 ON e2.event_id = ed.event_id" - " WHERE e.room_id = ? AND e.topological_ordering < ?" - " AND e2.topological_ordering >= ?", - (room_id, topological_ordering, topological_ordering) + "SELECT DISTINCT e.event_id FROM events_to_purge AS e" + " INNER JOIN event_edges AS ed ON e.event_id = ed.prev_event_id" + " INNER JOIN events AS e2 ON e2.event_id = ed.event_id" + " WHERE e2.topological_ordering >= ?", + (topological_ordering, ) ) new_backwards_extrems = txn.fetchall() - logger.debug("[purge] replacing backward extremities: %r", new_backwards_extrems) + logger.info("[purge] replacing backward extremities: %r", new_backwards_extrems) txn.execute( "DELETE FROM event_backward_extremities WHERE room_id = ?", @@ -2118,7 +1851,7 @@ class EventsStore(SQLBaseStore): ] ) - logger.debug("[purge] finding redundant state groups") + logger.info("[purge] finding redundant state groups") # Get all state groups that are only referenced by events that are # to be deleted. @@ -2126,24 +1859,23 @@ class EventsStore(SQLBaseStore): "SELECT state_group FROM event_to_state_groups" " INNER JOIN events USING (event_id)" " WHERE state_group IN (" - " SELECT DISTINCT state_group FROM events" + " SELECT DISTINCT state_group FROM events_to_purge" " INNER JOIN event_to_state_groups USING (event_id)" - " WHERE room_id = ? AND topological_ordering < ?" " )" " GROUP BY state_group HAVING MAX(topological_ordering) < ?", - (room_id, topological_ordering, topological_ordering) + (topological_ordering, ) ) state_rows = txn.fetchall() - logger.debug("[purge] found %i redundant state groups", len(state_rows)) + logger.info("[purge] found %i redundant state groups", len(state_rows)) # make a set of the redundant state groups, so that we can look them up # efficiently state_groups_to_delete = set([sg for sg, in state_rows]) # Now we get all the state groups that rely on these state groups - logger.debug("[purge] finding state groups which depend on redundant" - " state groups") + logger.info("[purge] finding state groups which depend on redundant" + " state groups") remaining_state_groups = [] for i in xrange(0, len(state_rows), 100): chunk = [sg for sg, in state_rows[i:i + 100]] @@ -2168,7 +1900,7 @@ class EventsStore(SQLBaseStore): # Now we turn the state groups that reference to-be-deleted state # groups to non delta versions. for sg in remaining_state_groups: - logger.debug("[purge] de-delta-ing remaining state group %s", sg) + logger.info("[purge] de-delta-ing remaining state group %s", sg) curr_state = self._get_state_groups_from_groups_txn( txn, [sg], types=None ) @@ -2205,7 +1937,7 @@ class EventsStore(SQLBaseStore): ], ) - logger.debug("[purge] removing redundant state groups") + logger.info("[purge] removing redundant state groups") txn.executemany( "DELETE FROM state_groups_state WHERE state_group = ?", state_rows @@ -2215,18 +1947,15 @@ class EventsStore(SQLBaseStore): state_rows ) - # Delete all non-state - logger.debug("[purge] removing events from event_to_state_groups") - txn.executemany( - "DELETE FROM event_to_state_groups WHERE event_id = ?", - [(event_id,) for event_id, _ in event_rows] - ) - - logger.debug("[purge] updating room_depth") + logger.info("[purge] removing events from event_to_state_groups") txn.execute( - "UPDATE room_depth SET min_depth = ? WHERE room_id = ?", - (topological_ordering, room_id,) + "DELETE FROM event_to_state_groups " + "WHERE event_id IN (SELECT event_id from events_to_purge)" ) + for event_id, _ in event_rows: + txn.call_after(self._get_state_group_for_event.invalidate, ( + event_id, + )) # Delete all remote non-state events for table in ( @@ -2238,28 +1967,60 @@ class EventsStore(SQLBaseStore): "event_edge_hashes", "event_edges", "event_forward_extremities", - "event_push_actions", "event_reference_hashes", "event_search", "event_signatures", "rejections", ): - logger.debug("[purge] removing remote non-state events from %s", table) + logger.info("[purge] removing events from %s", table) - txn.executemany( - "DELETE FROM %s WHERE event_id = ?" % (table,), - to_delete + txn.execute( + "DELETE FROM %s WHERE event_id IN (" + " SELECT event_id FROM events_to_purge WHERE should_delete" + ")" % (table,), + ) + + # event_push_actions lacks an index on event_id, and has one on + # (room_id, event_id) instead. + for table in ( + "event_push_actions", + ): + logger.info("[purge] removing events from %s", table) + + txn.execute( + "DELETE FROM %s WHERE room_id = ? AND event_id IN (" + " SELECT event_id FROM events_to_purge WHERE should_delete" + ")" % (table,), + (room_id, ) ) # Mark all state and own events as outliers - logger.debug("[purge] marking remaining events as outliers") - txn.executemany( + logger.info("[purge] marking remaining events as outliers") + txn.execute( "UPDATE events SET outlier = ?" - " WHERE event_id = ?", - [ - (True, event_id,) for event_id, state_key in event_rows - if state_key is not None or self.hs.is_mine_id(event_id) - ] + " WHERE event_id IN (" + " SELECT event_id FROM events_to_purge " + " WHERE NOT should_delete" + ")", + (True,), + ) + + # synapse tries to take out an exclusive lock on room_depth whenever it + # persists events (because upsert), and once we run this update, we + # will block that for the rest of our transaction. + # + # So, let's stick it at the end so that we don't block event + # persistence. + logger.info("[purge] updating room_depth") + txn.execute( + "UPDATE room_depth SET min_depth = ? WHERE room_id = ?", + (topological_ordering, room_id,) + ) + + # finally, drop the temp table. this will commit the txn in sqlite, + # so make sure to keep this actually last. + txn.execute( + "DROP TABLE events_to_purge" ) logger.info("[purge] done") @@ -2272,7 +2033,7 @@ class EventsStore(SQLBaseStore): to_2, so_2 = yield self._get_event_ordering(event_id2) defer.returnValue((to_1, so_1) > (to_2, so_2)) - @defer.inlineCallbacks + @cachedInlineCallbacks(max_entries=5000) def _get_event_ordering(self, event_id): res = yield self._simple_select_one( table="events", diff --git a/synapse/storage/events_worker.py b/synapse/storage/events_worker.py new file mode 100644 index 0000000000..2e23dd78ba --- /dev/null +++ b/synapse/storage/events_worker.py @@ -0,0 +1,395 @@ +# -*- coding: utf-8 -*- +# Copyright 2018 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. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. +from ._base import SQLBaseStore + +from twisted.internet import defer, reactor + +from synapse.events import FrozenEvent +from synapse.events.utils import prune_event + +from synapse.util.logcontext import ( + preserve_fn, PreserveLoggingContext, make_deferred_yieldable +) +from synapse.util.metrics import Measure +from synapse.api.errors import SynapseError + +from collections import namedtuple + +import logging +import simplejson as json + +# these are only included to make the type annotations work +from synapse.events import EventBase # noqa: F401 +from synapse.events.snapshot import EventContext # noqa: F401 + +logger = logging.getLogger(__name__) + + +# These values are used in the `enqueus_event` and `_do_fetch` methods to +# control how we batch/bulk fetch events from the database. +# The values are plucked out of thing air to make initial sync run faster +# on jki.re +# TODO: Make these configurable. +EVENT_QUEUE_THREADS = 3 # Max number of threads that will fetch events +EVENT_QUEUE_ITERATIONS = 3 # No. times we block waiting for requests for events +EVENT_QUEUE_TIMEOUT_S = 0.1 # Timeout when waiting for requests for events + + +_EventCacheEntry = namedtuple("_EventCacheEntry", ("event", "redacted_event")) + + +class EventsWorkerStore(SQLBaseStore): + + @defer.inlineCallbacks + def get_event(self, event_id, check_redacted=True, + get_prev_content=False, allow_rejected=False, + allow_none=False): + """Get an event from the database by event_id. + + Args: + event_id (str): The event_id of the event to fetch + check_redacted (bool): If True, check if event has been redacted + and redact it. + get_prev_content (bool): If True and event is a state event, + include the previous states content in the unsigned field. + allow_rejected (bool): If True return rejected events. + allow_none (bool): If True, return None if no event found, if + False throw an exception. + + Returns: + Deferred : A FrozenEvent. + """ + events = yield self._get_events( + [event_id], + check_redacted=check_redacted, + get_prev_content=get_prev_content, + allow_rejected=allow_rejected, + ) + + if not events and not allow_none: + raise SynapseError(404, "Could not find event %s" % (event_id,)) + + defer.returnValue(events[0] if events else None) + + @defer.inlineCallbacks + def get_events(self, event_ids, check_redacted=True, + get_prev_content=False, allow_rejected=False): + """Get events from the database + + Args: + event_ids (list): The event_ids of the events to fetch + check_redacted (bool): If True, check if event has been redacted + and redact it. + get_prev_content (bool): If True and event is a state event, + include the previous states content in the unsigned field. + allow_rejected (bool): If True return rejected events. + + Returns: + Deferred : Dict from event_id to event. + """ + events = yield self._get_events( + event_ids, + check_redacted=check_redacted, + get_prev_content=get_prev_content, + allow_rejected=allow_rejected, + ) + + defer.returnValue({e.event_id: e for e in events}) + + @defer.inlineCallbacks + def _get_events(self, event_ids, check_redacted=True, + get_prev_content=False, allow_rejected=False): + if not event_ids: + defer.returnValue([]) + + event_id_list = event_ids + event_ids = set(event_ids) + + event_entry_map = self._get_events_from_cache( + event_ids, + allow_rejected=allow_rejected, + ) + + missing_events_ids = [e for e in event_ids if e not in event_entry_map] + + if missing_events_ids: + missing_events = yield self._enqueue_events( + missing_events_ids, + check_redacted=check_redacted, + allow_rejected=allow_rejected, + ) + + event_entry_map.update(missing_events) + + events = [] + for event_id in event_id_list: + entry = event_entry_map.get(event_id, None) + if not entry: + continue + + if allow_rejected or not entry.event.rejected_reason: + if check_redacted and entry.redacted_event: + event = entry.redacted_event + else: + event = entry.event + + events.append(event) + + if get_prev_content: + if "replaces_state" in event.unsigned: + prev = yield self.get_event( + event.unsigned["replaces_state"], + get_prev_content=False, + allow_none=True, + ) + if prev: + event.unsigned = dict(event.unsigned) + event.unsigned["prev_content"] = prev.content + event.unsigned["prev_sender"] = prev.sender + + defer.returnValue(events) + + def _invalidate_get_event_cache(self, event_id): + self._get_event_cache.invalidate((event_id,)) + + def _get_events_from_cache(self, events, allow_rejected, update_metrics=True): + """Fetch events from the caches + + Args: + events (list(str)): list of event_ids to fetch + allow_rejected (bool): Whether to teturn events that were rejected + update_metrics (bool): Whether to update the cache hit ratio metrics + + Returns: + dict of event_id -> _EventCacheEntry for each event_id in cache. If + allow_rejected is `False` then there will still be an entry but it + will be `None` + """ + event_map = {} + + for event_id in events: + ret = self._get_event_cache.get( + (event_id,), None, + update_metrics=update_metrics, + ) + if not ret: + continue + + if allow_rejected or not ret.event.rejected_reason: + event_map[event_id] = ret + else: + event_map[event_id] = None + + return event_map + + def _do_fetch(self, conn): + """Takes a database connection and waits for requests for events from + the _event_fetch_list queue. + """ + event_list = [] + i = 0 + while True: + try: + with self._event_fetch_lock: + event_list = self._event_fetch_list + self._event_fetch_list = [] + + if not event_list: + single_threaded = self.database_engine.single_threaded + if single_threaded or i > EVENT_QUEUE_ITERATIONS: + self._event_fetch_ongoing -= 1 + return + else: + self._event_fetch_lock.wait(EVENT_QUEUE_TIMEOUT_S) + i += 1 + continue + i = 0 + + event_id_lists = zip(*event_list)[0] + event_ids = [ + item for sublist in event_id_lists for item in sublist + ] + + rows = self._new_transaction( + conn, "do_fetch", [], [], None, self._fetch_event_rows, event_ids + ) + + row_dict = { + r["event_id"]: r + for r in rows + } + + # We only want to resolve deferreds from the main thread + def fire(lst, res): + for ids, d in lst: + if not d.called: + try: + with PreserveLoggingContext(): + d.callback([ + res[i] + for i in ids + if i in res + ]) + except Exception: + logger.exception("Failed to callback") + with PreserveLoggingContext(): + reactor.callFromThread(fire, event_list, row_dict) + except Exception as e: + logger.exception("do_fetch") + + # We only want to resolve deferreds from the main thread + def fire(evs): + for _, d in evs: + if not d.called: + with PreserveLoggingContext(): + d.errback(e) + + if event_list: + with PreserveLoggingContext(): + reactor.callFromThread(fire, event_list) + + @defer.inlineCallbacks + def _enqueue_events(self, events, check_redacted=True, allow_rejected=False): + """Fetches events from the database using the _event_fetch_list. This + allows batch and bulk fetching of events - it allows us to fetch events + without having to create a new transaction for each request for events. + """ + if not events: + defer.returnValue({}) + + events_d = defer.Deferred() + with self._event_fetch_lock: + self._event_fetch_list.append( + (events, events_d) + ) + + self._event_fetch_lock.notify() + + if self._event_fetch_ongoing < EVENT_QUEUE_THREADS: + self._event_fetch_ongoing += 1 + should_start = True + else: + should_start = False + + if should_start: + with PreserveLoggingContext(): + self.runWithConnection( + self._do_fetch + ) + + logger.debug("Loading %d events", len(events)) + with PreserveLoggingContext(): + rows = yield events_d + logger.debug("Loaded %d events (%d rows)", len(events), len(rows)) + + if not allow_rejected: + rows[:] = [r for r in rows if not r["rejects"]] + + res = yield make_deferred_yieldable(defer.gatherResults( + [ + preserve_fn(self._get_event_from_row)( + row["internal_metadata"], row["json"], row["redacts"], + rejected_reason=row["rejects"], + ) + for row in rows + ], + consumeErrors=True + )) + + defer.returnValue({ + e.event.event_id: e + for e in res if e + }) + + def _fetch_event_rows(self, txn, events): + rows = [] + N = 200 + for i in range(1 + len(events) / N): + evs = events[i * N:(i + 1) * N] + if not evs: + break + + sql = ( + "SELECT " + " e.event_id as event_id, " + " e.internal_metadata," + " e.json," + " r.redacts as redacts," + " rej.event_id as rejects " + " FROM event_json as e" + " LEFT JOIN rejections as rej USING (event_id)" + " LEFT JOIN redactions as r ON e.event_id = r.redacts" + " WHERE e.event_id IN (%s)" + ) % (",".join(["?"] * len(evs)),) + + txn.execute(sql, evs) + rows.extend(self.cursor_to_dict(txn)) + + return rows + + @defer.inlineCallbacks + def _get_event_from_row(self, internal_metadata, js, redacted, + rejected_reason=None): + with Measure(self._clock, "_get_event_from_row"): + d = json.loads(js) + internal_metadata = json.loads(internal_metadata) + + if rejected_reason: + rejected_reason = yield self._simple_select_one_onecol( + table="rejections", + keyvalues={"event_id": rejected_reason}, + retcol="reason", + desc="_get_event_from_row_rejected_reason", + ) + + original_ev = FrozenEvent( + d, + internal_metadata_dict=internal_metadata, + rejected_reason=rejected_reason, + ) + + redacted_event = None + if redacted: + redacted_event = prune_event(original_ev) + + redaction_id = yield self._simple_select_one_onecol( + table="redactions", + keyvalues={"redacts": redacted_event.event_id}, + retcol="event_id", + desc="_get_event_from_row_redactions", + ) + + redacted_event.unsigned["redacted_by"] = redaction_id + # Get the redaction event. + + because = yield self.get_event( + redaction_id, + check_redacted=False, + allow_none=True, + ) + + if because: + # It's fine to do add the event directly, since get_pdu_json + # will serialise this field correctly + redacted_event.unsigned["redacted_because"] = because + + cache_entry = _EventCacheEntry( + event=original_ev, + redacted_event=redacted_event, + ) + + self._get_event_cache.prefill((original_ev.event_id,), cache_entry) + + defer.returnValue(cache_entry) diff --git a/synapse/storage/media_repository.py b/synapse/storage/media_repository.py index a66ff7c1e0..e6cdbb0545 100644 --- a/synapse/storage/media_repository.py +++ b/synapse/storage/media_repository.py @@ -29,9 +29,6 @@ class MediaRepositoryStore(BackgroundUpdateStore): where_clause='url_cache IS NOT NULL', ) - def get_default_thumbnails(self, top_level_type, sub_type): - return [] - def get_local_media(self, media_id): """Get the metadata for a local piece of media Returns: @@ -176,7 +173,14 @@ class MediaRepositoryStore(BackgroundUpdateStore): desc="store_cached_remote_media", ) - def update_cached_last_access_time(self, origin_id_tuples, time_ts): + def update_cached_last_access_time(self, local_media, remote_media, time_ms): + """Updates the last access time of the given media + + Args: + local_media (iterable[str]): Set of media_ids + remote_media (iterable[(str, str)]): Set of (server_name, media_id) + time_ms: Current time in milliseconds + """ def update_cache_txn(txn): sql = ( "UPDATE remote_media_cache SET last_access_ts = ?" @@ -184,8 +188,18 @@ class MediaRepositoryStore(BackgroundUpdateStore): ) txn.executemany(sql, ( - (time_ts, media_origin, media_id) - for media_origin, media_id in origin_id_tuples + (time_ms, media_origin, media_id) + for media_origin, media_id in remote_media + )) + + sql = ( + "UPDATE local_media_repository SET last_access_ts = ?" + " WHERE media_id = ?" + ) + + txn.executemany(sql, ( + (time_ms, media_id) + for media_id in local_media )) return self.runInteraction("update_cached_last_access_time", update_cache_txn) diff --git a/synapse/storage/prepare_database.py b/synapse/storage/prepare_database.py index d1691bbac2..c845a0cec5 100644 --- a/synapse/storage/prepare_database.py +++ b/synapse/storage/prepare_database.py @@ -25,7 +25,7 @@ logger = logging.getLogger(__name__) # Remember to update this number every time a change is made to database # schema files, so the users will be informed on server restarts. -SCHEMA_VERSION = 46 +SCHEMA_VERSION = 47 dir_path = os.path.abspath(os.path.dirname(__file__)) diff --git a/synapse/storage/profile.py b/synapse/storage/profile.py index ec02e73bc2..8612bd5ecc 100644 --- a/synapse/storage/profile.py +++ b/synapse/storage/profile.py @@ -21,14 +21,7 @@ from synapse.api.errors import StoreError from ._base import SQLBaseStore -class ProfileStore(SQLBaseStore): - def create_profile(self, user_localpart): - return self._simple_insert( - table="profiles", - values={"user_id": user_localpart}, - desc="create_profile", - ) - +class ProfileWorkerStore(SQLBaseStore): @defer.inlineCallbacks def get_profileinfo(self, user_localpart): try: @@ -61,14 +54,6 @@ class ProfileStore(SQLBaseStore): desc="get_profile_displayname", ) - def set_profile_displayname(self, user_localpart, new_displayname): - return self._simple_update_one( - table="profiles", - keyvalues={"user_id": user_localpart}, - updatevalues={"displayname": new_displayname}, - desc="set_profile_displayname", - ) - def get_profile_avatar_url(self, user_localpart): return self._simple_select_one_onecol( table="profiles", @@ -77,14 +62,6 @@ class ProfileStore(SQLBaseStore): desc="get_profile_avatar_url", ) - def set_profile_avatar_url(self, user_localpart, new_avatar_url): - return self._simple_update_one( - table="profiles", - keyvalues={"user_id": user_localpart}, - updatevalues={"avatar_url": new_avatar_url}, - desc="set_profile_avatar_url", - ) - def get_from_remote_profile_cache(self, user_id): return self._simple_select_one( table="remote_profile_cache", @@ -94,6 +71,31 @@ class ProfileStore(SQLBaseStore): desc="get_from_remote_profile_cache", ) + +class ProfileStore(ProfileWorkerStore): + def create_profile(self, user_localpart): + return self._simple_insert( + table="profiles", + values={"user_id": user_localpart}, + desc="create_profile", + ) + + def set_profile_displayname(self, user_localpart, new_displayname): + return self._simple_update_one( + table="profiles", + keyvalues={"user_id": user_localpart}, + updatevalues={"displayname": new_displayname}, + desc="set_profile_displayname", + ) + + def set_profile_avatar_url(self, user_localpart, new_avatar_url): + return self._simple_update_one( + table="profiles", + keyvalues={"user_id": user_localpart}, + updatevalues={"avatar_url": new_avatar_url}, + desc="set_profile_avatar_url", + ) + def add_remote_profile_cache(self, user_id, displayname, avatar_url): """Ensure we are caching the remote user's profiles. diff --git a/synapse/storage/push_rule.py b/synapse/storage/push_rule.py index 8758b1c0c7..04a0b59a39 100644 --- a/synapse/storage/push_rule.py +++ b/synapse/storage/push_rule.py @@ -1,5 +1,6 @@ # -*- coding: utf-8 -*- # Copyright 2014-2016 OpenMarket Ltd +# Copyright 2018 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. @@ -14,11 +15,17 @@ # limitations under the License. from ._base import SQLBaseStore +from synapse.storage.appservice import ApplicationServiceWorkerStore +from synapse.storage.pusher import PusherWorkerStore +from synapse.storage.receipts import ReceiptsWorkerStore +from synapse.storage.roommember import RoomMemberWorkerStore from synapse.util.caches.descriptors import cachedInlineCallbacks, cachedList +from synapse.util.caches.stream_change_cache import StreamChangeCache from synapse.push.baserules import list_with_base_rules from synapse.api.constants import EventTypes from twisted.internet import defer +import abc import logging import simplejson as json @@ -48,7 +55,43 @@ def _load_rules(rawrules, enabled_map): return rules -class PushRuleStore(SQLBaseStore): +class PushRulesWorkerStore(ApplicationServiceWorkerStore, + ReceiptsWorkerStore, + PusherWorkerStore, + RoomMemberWorkerStore, + SQLBaseStore): + """This is an abstract base class where subclasses must implement + `get_max_push_rules_stream_id` which can be called in the initializer. + """ + + # This ABCMeta metaclass ensures that we cannot be instantiated without + # the abstract methods being implemented. + __metaclass__ = abc.ABCMeta + + def __init__(self, db_conn, hs): + super(PushRulesWorkerStore, self).__init__(db_conn, hs) + + push_rules_prefill, push_rules_id = self._get_cache_dict( + db_conn, "push_rules_stream", + entity_column="user_id", + stream_column="stream_id", + max_value=self.get_max_push_rules_stream_id(), + ) + + self.push_rules_stream_cache = StreamChangeCache( + "PushRulesStreamChangeCache", push_rules_id, + prefilled_cache=push_rules_prefill, + ) + + @abc.abstractmethod + def get_max_push_rules_stream_id(self): + """Get the position of the push rules stream. + + Returns: + int + """ + raise NotImplementedError() + @cachedInlineCallbacks(max_entries=5000) def get_push_rules_for_user(self, user_id): rows = yield self._simple_select_list( @@ -89,6 +132,22 @@ class PushRuleStore(SQLBaseStore): r['rule_id']: False if r['enabled'] == 0 else True for r in results }) + def have_push_rules_changed_for_user(self, user_id, last_id): + if not self.push_rules_stream_cache.has_entity_changed(user_id, last_id): + return defer.succeed(False) + else: + def have_push_rules_changed_txn(txn): + sql = ( + "SELECT COUNT(stream_id) FROM push_rules_stream" + " WHERE user_id = ? AND ? < stream_id" + ) + txn.execute(sql, (user_id, last_id)) + count, = txn.fetchone() + return bool(count) + return self.runInteraction( + "have_push_rules_changed", have_push_rules_changed_txn + ) + @cachedList(cached_method_name="get_push_rules_for_user", list_name="user_ids", num_args=1, inlineCallbacks=True) def bulk_get_push_rules(self, user_ids): @@ -228,6 +287,8 @@ class PushRuleStore(SQLBaseStore): results.setdefault(row['user_name'], {})[row['rule_id']] = enabled defer.returnValue(results) + +class PushRuleStore(PushRulesWorkerStore): @defer.inlineCallbacks def add_push_rule( self, user_id, rule_id, priority_class, conditions, actions, @@ -526,21 +587,8 @@ class PushRuleStore(SQLBaseStore): room stream ordering it corresponds to.""" return self._push_rules_stream_id_gen.get_current_token() - def have_push_rules_changed_for_user(self, user_id, last_id): - if not self.push_rules_stream_cache.has_entity_changed(user_id, last_id): - return defer.succeed(False) - else: - def have_push_rules_changed_txn(txn): - sql = ( - "SELECT COUNT(stream_id) FROM push_rules_stream" - " WHERE user_id = ? AND ? < stream_id" - ) - txn.execute(sql, (user_id, last_id)) - count, = txn.fetchone() - return bool(count) - return self.runInteraction( - "have_push_rules_changed", have_push_rules_changed_txn - ) + def get_max_push_rules_stream_id(self): + return self.get_push_rules_stream_token()[0] class RuleNotFoundException(Exception): diff --git a/synapse/storage/pusher.py b/synapse/storage/pusher.py index 3d8b4d5d5b..307660b99a 100644 --- a/synapse/storage/pusher.py +++ b/synapse/storage/pusher.py @@ -1,5 +1,6 @@ # -*- coding: utf-8 -*- # Copyright 2014-2016 OpenMarket Ltd +# Copyright 2018 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. @@ -27,7 +28,7 @@ import types logger = logging.getLogger(__name__) -class PusherStore(SQLBaseStore): +class PusherWorkerStore(SQLBaseStore): def _decode_pushers_rows(self, rows): for r in rows: dataJson = r['data'] @@ -102,9 +103,6 @@ class PusherStore(SQLBaseStore): rows = yield self.runInteraction("get_all_pushers", get_pushers) defer.returnValue(rows) - def get_pushers_stream_token(self): - return self._pushers_id_gen.get_current_token() - def get_all_updated_pushers(self, last_id, current_id, limit): if last_id == current_id: return defer.succeed(([], [])) @@ -198,6 +196,11 @@ class PusherStore(SQLBaseStore): defer.returnValue(result) + +class PusherStore(PusherWorkerStore): + def get_pushers_stream_token(self): + return self._pushers_id_gen.get_current_token() + @defer.inlineCallbacks def add_pusher(self, user_id, access_token, kind, app_id, app_display_name, device_display_name, @@ -230,14 +233,18 @@ class PusherStore(SQLBaseStore): ) 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,) + self.runInteraction( + "add_pusher", + self._invalidate_cache_and_stream, + self.get_if_user_has_pusher, (user_id,) + ) @defer.inlineCallbacks def delete_pusher_by_app_id_pushkey_user_id(self, app_id, pushkey, user_id): def delete_pusher_txn(txn, stream_id): - txn.call_after(self.get_if_user_has_pusher.invalidate, (user_id,)) + self._invalidate_cache_and_stream( + txn, self.get_if_user_has_pusher, (user_id,) + ) self._simple_delete_one_txn( txn, diff --git a/synapse/storage/receipts.py b/synapse/storage/receipts.py index 2c3aa33693..63997ed449 100644 --- a/synapse/storage/receipts.py +++ b/synapse/storage/receipts.py @@ -1,5 +1,6 @@ # -*- coding: utf-8 -*- # Copyright 2014-2016 OpenMarket Ltd +# Copyright 2018 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. @@ -14,11 +15,13 @@ # limitations under the License. from ._base import SQLBaseStore +from .util.id_generators import StreamIdGenerator from synapse.util.caches.descriptors import cachedInlineCallbacks, cachedList, cached from synapse.util.caches.stream_change_cache import StreamChangeCache from twisted.internet import defer +import abc import logging import simplejson as json @@ -26,39 +29,36 @@ import simplejson as json logger = logging.getLogger(__name__) -class ReceiptsStore(SQLBaseStore): +class ReceiptsWorkerStore(SQLBaseStore): + """This is an abstract base class where subclasses must implement + `get_max_receipt_stream_id` which can be called in the initializer. + """ + + # This ABCMeta metaclass ensures that we cannot be instantiated without + # the abstract methods being implemented. + __metaclass__ = abc.ABCMeta + def __init__(self, db_conn, hs): - super(ReceiptsStore, self).__init__(db_conn, hs) + super(ReceiptsWorkerStore, self).__init__(db_conn, hs) self._receipts_stream_cache = StreamChangeCache( - "ReceiptsRoomChangeCache", self._receipts_id_gen.get_current_token() + "ReceiptsRoomChangeCache", self.get_max_receipt_stream_id() ) + @abc.abstractmethod + def get_max_receipt_stream_id(self): + """Get the current max stream ID for receipts stream + + Returns: + int + """ + raise NotImplementedError() + @cachedInlineCallbacks() def get_users_with_read_receipts_in_room(self, room_id): receipts = yield self.get_receipts_for_room(room_id, "m.read") defer.returnValue(set(r['user_id'] for r in receipts)) - def _invalidate_get_users_with_receipts_in_room(self, room_id, receipt_type, - user_id): - if receipt_type != "m.read": - return - - # Returns an ObservableDeferred - res = self.get_users_with_read_receipts_in_room.cache.get( - room_id, None, update_metrics=False, - ) - - if res: - if isinstance(res, defer.Deferred) and res.called: - res = res.result - if user_id in res: - # We'd only be adding to the set, so no point invalidating if the - # user is already there - return - - self.get_users_with_read_receipts_in_room.invalidate((room_id,)) - @cached(num_args=2) def get_receipts_for_room(self, room_id, receipt_type): return self._simple_select_list( @@ -270,6 +270,59 @@ class ReceiptsStore(SQLBaseStore): } defer.returnValue(results) + def get_all_updated_receipts(self, last_id, current_id, limit=None): + if last_id == current_id: + return defer.succeed([]) + + def get_all_updated_receipts_txn(txn): + sql = ( + "SELECT stream_id, room_id, receipt_type, user_id, event_id, data" + " FROM receipts_linearized" + " WHERE ? < stream_id AND stream_id <= ?" + " ORDER BY stream_id ASC" + ) + args = [last_id, current_id] + if limit is not None: + sql += " LIMIT ?" + args.append(limit) + txn.execute(sql, args) + + return txn.fetchall() + return self.runInteraction( + "get_all_updated_receipts", get_all_updated_receipts_txn + ) + + def _invalidate_get_users_with_receipts_in_room(self, room_id, receipt_type, + user_id): + if receipt_type != "m.read": + return + + # Returns an ObservableDeferred + res = self.get_users_with_read_receipts_in_room.cache.get( + room_id, None, update_metrics=False, + ) + + if res: + if isinstance(res, defer.Deferred) and res.called: + res = res.result + if user_id in res: + # We'd only be adding to the set, so no point invalidating if the + # user is already there + return + + self.get_users_with_read_receipts_in_room.invalidate((room_id,)) + + +class ReceiptsStore(ReceiptsWorkerStore): + def __init__(self, db_conn, hs): + # We instantiate this first as the ReceiptsWorkerStore constructor + # needs to be able to call get_max_receipt_stream_id + self._receipts_id_gen = StreamIdGenerator( + db_conn, "receipts_linearized", "stream_id" + ) + + super(ReceiptsStore, self).__init__(db_conn, hs) + def get_max_receipt_stream_id(self): return self._receipts_id_gen.get_current_token() @@ -457,25 +510,3 @@ class ReceiptsStore(SQLBaseStore): "data": json.dumps(data), } ) - - def get_all_updated_receipts(self, last_id, current_id, limit=None): - if last_id == current_id: - return defer.succeed([]) - - def get_all_updated_receipts_txn(txn): - sql = ( - "SELECT stream_id, room_id, receipt_type, user_id, event_id, data" - " FROM receipts_linearized" - " WHERE ? < stream_id AND stream_id <= ?" - " ORDER BY stream_id ASC" - ) - args = [last_id, current_id] - if limit is not None: - sql += " LIMIT ?" - args.append(limit) - txn.execute(sql, args) - - return txn.fetchall() - return self.runInteraction( - "get_all_updated_receipts", get_all_updated_receipts_txn - ) diff --git a/synapse/storage/registration.py b/synapse/storage/registration.py index 3aa810981f..d809b2ba46 100644 --- a/synapse/storage/registration.py +++ b/synapse/storage/registration.py @@ -19,10 +19,70 @@ from twisted.internet import defer from synapse.api.errors import StoreError, Codes from synapse.storage import background_updates +from synapse.storage._base import SQLBaseStore from synapse.util.caches.descriptors import cached, cachedInlineCallbacks -class RegistrationStore(background_updates.BackgroundUpdateStore): +class RegistrationWorkerStore(SQLBaseStore): + @cached() + def get_user_by_id(self, user_id): + return self._simple_select_one( + table="users", + keyvalues={ + "name": user_id, + }, + retcols=["name", "password_hash", "is_guest"], + allow_none=True, + desc="get_user_by_id", + ) + + @cached() + def get_user_by_access_token(self, token): + """Get a user from the given access token. + + Args: + token (str): The access token of a user. + Returns: + defer.Deferred: None, if the token did not match, otherwise dict + including the keys `name`, `is_guest`, `device_id`, `token_id`. + """ + return self.runInteraction( + "get_user_by_access_token", + self._query_for_auth, + token + ) + + @defer.inlineCallbacks + def is_server_admin(self, user): + res = yield self._simple_select_one_onecol( + table="users", + keyvalues={"name": user.to_string()}, + retcol="admin", + allow_none=True, + desc="is_server_admin", + ) + + defer.returnValue(res if res else False) + + def _query_for_auth(self, txn, token): + sql = ( + "SELECT users.name, users.is_guest, access_tokens.id as token_id," + " access_tokens.device_id" + " FROM users" + " INNER JOIN access_tokens on users.name = access_tokens.user_id" + " WHERE token = ?" + ) + + txn.execute(sql, (token,)) + rows = self.cursor_to_dict(txn) + if rows: + return rows[0] + + return None + + +class RegistrationStore(RegistrationWorkerStore, + background_updates.BackgroundUpdateStore): def __init__(self, db_conn, hs): super(RegistrationStore, self).__init__(db_conn, hs) @@ -39,12 +99,7 @@ class RegistrationStore(background_updates.BackgroundUpdateStore): # we no longer use refresh tokens, but it's possible that some people # might have a background update queued to build this index. Just # clear the background update. - @defer.inlineCallbacks - def noop_update(progress, batch_size): - yield self._end_background_update("refresh_tokens_device_index") - defer.returnValue(1) - self.register_background_update_handler( - "refresh_tokens_device_index", noop_update) + self.register_noop_background_update("refresh_tokens_device_index") @defer.inlineCallbacks def add_access_token_to_user(self, user_id, token, device_id=None): @@ -192,18 +247,6 @@ class RegistrationStore(background_updates.BackgroundUpdateStore): ) txn.call_after(self.is_guest.invalidate, (user_id,)) - @cached() - def get_user_by_id(self, user_id): - return self._simple_select_one( - table="users", - keyvalues={ - "name": user_id, - }, - retcols=["name", "password_hash", "is_guest"], - allow_none=True, - desc="get_user_by_id", - ) - def get_users_by_id_case_insensitive(self, user_id): """Gets users that match user_id case insensitively. Returns a mapping of user_id -> password_hash. @@ -309,34 +352,6 @@ class RegistrationStore(background_updates.BackgroundUpdateStore): return self.runInteraction("delete_access_token", f) - @cached() - def get_user_by_access_token(self, token): - """Get a user from the given access token. - - Args: - token (str): The access token of a user. - Returns: - defer.Deferred: None, if the token did not match, otherwise dict - including the keys `name`, `is_guest`, `device_id`, `token_id`. - """ - return self.runInteraction( - "get_user_by_access_token", - self._query_for_auth, - token - ) - - @defer.inlineCallbacks - def is_server_admin(self, user): - res = yield self._simple_select_one_onecol( - table="users", - keyvalues={"name": user.to_string()}, - retcol="admin", - allow_none=True, - desc="is_server_admin", - ) - - defer.returnValue(res if res else False) - @cachedInlineCallbacks() def is_guest(self, user_id): res = yield self._simple_select_one_onecol( @@ -349,22 +364,6 @@ class RegistrationStore(background_updates.BackgroundUpdateStore): defer.returnValue(res if res else False) - def _query_for_auth(self, txn, token): - sql = ( - "SELECT users.name, users.is_guest, access_tokens.id as token_id," - " access_tokens.device_id" - " FROM users" - " INNER JOIN access_tokens on users.name = access_tokens.user_id" - " WHERE token = ?" - ) - - txn.execute(sql, (token,)) - rows = self.cursor_to_dict(txn) - if rows: - return rows[0] - - return None - @defer.inlineCallbacks def user_add_threepid(self, user_id, medium, address, validated_at, added_at): yield self._simple_upsert("user_threepids", { diff --git a/synapse/storage/room.py b/synapse/storage/room.py index 2051d8506d..908551d6d9 100644 --- a/synapse/storage/room.py +++ b/synapse/storage/room.py @@ -16,11 +16,10 @@ from twisted.internet import defer from synapse.api.errors import StoreError +from synapse.storage._base import SQLBaseStore +from synapse.storage.search import SearchStore from synapse.util.caches.descriptors import cached, cachedInlineCallbacks -from ._base import SQLBaseStore -from .engines import PostgresEngine, Sqlite3Engine - import collections import logging import simplejson as json @@ -40,7 +39,138 @@ RatelimitOverride = collections.namedtuple( ) -class RoomStore(SQLBaseStore): +class RoomWorkerStore(SQLBaseStore): + def get_public_room_ids(self): + return self._simple_select_onecol( + table="rooms", + keyvalues={ + "is_public": True, + }, + retcol="room_id", + desc="get_public_room_ids", + ) + + @cached(num_args=2, max_entries=100) + def get_public_room_ids_at_stream_id(self, stream_id, network_tuple): + """Get pulbic rooms for a particular list, or across all lists. + + Args: + stream_id (int) + network_tuple (ThirdPartyInstanceID): The list to use (None, None) + means the main list, None means all lsits. + """ + return self.runInteraction( + "get_public_room_ids_at_stream_id", + self.get_public_room_ids_at_stream_id_txn, + stream_id, network_tuple=network_tuple + ) + + def get_public_room_ids_at_stream_id_txn(self, txn, stream_id, + network_tuple): + return { + rm + for rm, vis in self.get_published_at_stream_id_txn( + txn, stream_id, network_tuple=network_tuple + ).items() + if vis + } + + def get_published_at_stream_id_txn(self, txn, stream_id, network_tuple): + if network_tuple: + # We want to get from a particular list. No aggregation required. + + sql = (""" + SELECT room_id, visibility FROM public_room_list_stream + INNER JOIN ( + SELECT room_id, max(stream_id) AS stream_id + FROM public_room_list_stream + WHERE stream_id <= ? %s + GROUP BY room_id + ) grouped USING (room_id, stream_id) + """) + + if network_tuple.appservice_id is not None: + txn.execute( + sql % ("AND appservice_id = ? AND network_id = ?",), + (stream_id, network_tuple.appservice_id, network_tuple.network_id,) + ) + else: + txn.execute( + sql % ("AND appservice_id IS NULL",), + (stream_id,) + ) + return dict(txn) + else: + # We want to get from all lists, so we need to aggregate the results + + logger.info("Executing full list") + + sql = (""" + SELECT room_id, visibility + FROM public_room_list_stream + INNER JOIN ( + SELECT + room_id, max(stream_id) AS stream_id, appservice_id, + network_id + FROM public_room_list_stream + WHERE stream_id <= ? + GROUP BY room_id, appservice_id, network_id + ) grouped USING (room_id, stream_id) + """) + + txn.execute( + sql, + (stream_id,) + ) + + results = {} + # A room is visible if its visible on any list. + for room_id, visibility in txn: + results[room_id] = bool(visibility) or results.get(room_id, False) + + return results + + def get_public_room_changes(self, prev_stream_id, new_stream_id, + network_tuple): + def get_public_room_changes_txn(txn): + then_rooms = self.get_public_room_ids_at_stream_id_txn( + txn, prev_stream_id, network_tuple + ) + + now_rooms_dict = self.get_published_at_stream_id_txn( + txn, new_stream_id, network_tuple + ) + + now_rooms_visible = set( + rm for rm, vis in now_rooms_dict.items() if vis + ) + now_rooms_not_visible = set( + rm for rm, vis in now_rooms_dict.items() if not vis + ) + + newly_visible = now_rooms_visible - then_rooms + newly_unpublished = now_rooms_not_visible & then_rooms + + return newly_visible, newly_unpublished + + return self.runInteraction( + "get_public_room_changes", get_public_room_changes_txn + ) + + @cached(max_entries=10000) + def is_room_blocked(self, room_id): + return self._simple_select_one_onecol( + table="blocked_rooms", + keyvalues={ + "room_id": room_id, + }, + retcol="1", + allow_none=True, + desc="is_room_blocked", + ) + + +class RoomStore(RoomWorkerStore, SearchStore): @defer.inlineCallbacks def store_room(self, room_id, room_creator_user_id, is_public): @@ -227,16 +357,6 @@ class RoomStore(SQLBaseStore): ) self.hs.get_notifier().on_new_replication_data() - def get_public_room_ids(self): - return self._simple_select_onecol( - table="rooms", - keyvalues={ - "is_public": True, - }, - retcol="room_id", - desc="get_public_room_ids", - ) - def get_room_count(self): """Retrieve a list of all rooms """ @@ -263,8 +383,8 @@ class RoomStore(SQLBaseStore): }, ) - self._store_event_search_txn( - txn, event, "content.topic", event.content["topic"] + self.store_event_search_txn( + txn, event, "content.topic", event.content["topic"], ) def _store_room_name_txn(self, txn, event): @@ -279,14 +399,14 @@ class RoomStore(SQLBaseStore): } ) - self._store_event_search_txn( - txn, event, "content.name", event.content["name"] + self.store_event_search_txn( + txn, event, "content.name", event.content["name"], ) def _store_room_message_txn(self, txn, event): if hasattr(event, "content") and "body" in event.content: - self._store_event_search_txn( - txn, event, "content.body", event.content["body"] + self.store_event_search_txn( + txn, event, "content.body", event.content["body"], ) def _store_history_visibility_txn(self, txn, event): @@ -308,31 +428,6 @@ class RoomStore(SQLBaseStore): event.content[key] )) - def _store_event_search_txn(self, txn, event, key, value): - if isinstance(self.database_engine, PostgresEngine): - sql = ( - "INSERT INTO event_search" - " (event_id, room_id, key, vector, stream_ordering, origin_server_ts)" - " VALUES (?,?,?,to_tsvector('english', ?),?,?)" - ) - txn.execute( - sql, - ( - event.event_id, event.room_id, key, value, - event.internal_metadata.stream_ordering, - event.origin_server_ts, - ) - ) - elif isinstance(self.database_engine, Sqlite3Engine): - sql = ( - "INSERT INTO event_search (event_id, room_id, key, value)" - " VALUES (?,?,?,?)" - ) - txn.execute(sql, (event.event_id, event.room_id, key, value,)) - else: - # This should be unreachable. - raise Exception("Unrecognized database engine") - def add_event_report(self, room_id, event_id, user_id, reason, content, received_ts): next_id = self._event_reports_id_gen.get_next() @@ -353,113 +448,6 @@ class RoomStore(SQLBaseStore): def get_current_public_room_stream_id(self): return self._public_room_id_gen.get_current_token() - @cached(num_args=2, max_entries=100) - def get_public_room_ids_at_stream_id(self, stream_id, network_tuple): - """Get pulbic rooms for a particular list, or across all lists. - - Args: - stream_id (int) - network_tuple (ThirdPartyInstanceID): The list to use (None, None) - means the main list, None means all lsits. - """ - return self.runInteraction( - "get_public_room_ids_at_stream_id", - self.get_public_room_ids_at_stream_id_txn, - stream_id, network_tuple=network_tuple - ) - - def get_public_room_ids_at_stream_id_txn(self, txn, stream_id, - network_tuple): - return { - rm - for rm, vis in self.get_published_at_stream_id_txn( - txn, stream_id, network_tuple=network_tuple - ).items() - if vis - } - - def get_published_at_stream_id_txn(self, txn, stream_id, network_tuple): - if network_tuple: - # We want to get from a particular list. No aggregation required. - - sql = (""" - SELECT room_id, visibility FROM public_room_list_stream - INNER JOIN ( - SELECT room_id, max(stream_id) AS stream_id - FROM public_room_list_stream - WHERE stream_id <= ? %s - GROUP BY room_id - ) grouped USING (room_id, stream_id) - """) - - if network_tuple.appservice_id is not None: - txn.execute( - sql % ("AND appservice_id = ? AND network_id = ?",), - (stream_id, network_tuple.appservice_id, network_tuple.network_id,) - ) - else: - txn.execute( - sql % ("AND appservice_id IS NULL",), - (stream_id,) - ) - return dict(txn) - else: - # We want to get from all lists, so we need to aggregate the results - - logger.info("Executing full list") - - sql = (""" - SELECT room_id, visibility - FROM public_room_list_stream - INNER JOIN ( - SELECT - room_id, max(stream_id) AS stream_id, appservice_id, - network_id - FROM public_room_list_stream - WHERE stream_id <= ? - GROUP BY room_id, appservice_id, network_id - ) grouped USING (room_id, stream_id) - """) - - txn.execute( - sql, - (stream_id,) - ) - - results = {} - # A room is visible if its visible on any list. - for room_id, visibility in txn: - results[room_id] = bool(visibility) or results.get(room_id, False) - - return results - - def get_public_room_changes(self, prev_stream_id, new_stream_id, - network_tuple): - def get_public_room_changes_txn(txn): - then_rooms = self.get_public_room_ids_at_stream_id_txn( - txn, prev_stream_id, network_tuple - ) - - now_rooms_dict = self.get_published_at_stream_id_txn( - txn, new_stream_id, network_tuple - ) - - now_rooms_visible = set( - rm for rm, vis in now_rooms_dict.items() if vis - ) - now_rooms_not_visible = set( - rm for rm, vis in now_rooms_dict.items() if not vis - ) - - newly_visible = now_rooms_visible - then_rooms - newly_unpublished = now_rooms_not_visible & then_rooms - - return newly_visible, newly_unpublished - - return self.runInteraction( - "get_public_room_changes", get_public_room_changes_txn - ) - def get_all_new_public_rooms(self, prev_id, current_id, limit): def get_all_new_public_rooms(txn): sql = (""" @@ -509,18 +497,6 @@ class RoomStore(SQLBaseStore): else: defer.returnValue(None) - @cached(max_entries=10000) - def is_room_blocked(self, room_id): - return self._simple_select_one_onecol( - table="blocked_rooms", - keyvalues={ - "room_id": room_id, - }, - retcol="1", - allow_none=True, - desc="is_room_blocked", - ) - @defer.inlineCallbacks def block_room(self, room_id, user_id): yield self._simple_insert( @@ -531,75 +507,120 @@ class RoomStore(SQLBaseStore): }, desc="block_room", ) - self.is_room_blocked.invalidate((room_id,)) + yield self.runInteraction( + "block_room_invalidation", + self._invalidate_cache_and_stream, + self.is_room_blocked, (room_id,), + ) + + def get_media_mxcs_in_room(self, room_id): + """Retrieves all the local and remote media MXC URIs in a given room + + Args: + room_id (str) + + Returns: + The local and remote media as a lists of tuples where the key is + the hostname and the value is the media ID. + """ + def _get_media_mxcs_in_room_txn(txn): + local_mxcs, remote_mxcs = self._get_media_mxcs_in_room_txn(txn, room_id) + local_media_mxcs = [] + remote_media_mxcs = [] + + # Convert the IDs to MXC URIs + for media_id in local_mxcs: + local_media_mxcs.append("mxc://%s/%s" % (self.hostname, media_id)) + for hostname, media_id in remote_mxcs: + remote_media_mxcs.append("mxc://%s/%s" % (hostname, media_id)) + + return local_media_mxcs, remote_media_mxcs + return self.runInteraction("get_media_ids_in_room", _get_media_mxcs_in_room_txn) def quarantine_media_ids_in_room(self, room_id, quarantined_by): """For a room loops through all events with media and quarantines the associated media """ - def _get_media_ids_in_room(txn): - mxc_re = re.compile("^mxc://([^/]+)/([^/#?]+)") + def _quarantine_media_in_room_txn(txn): + local_mxcs, remote_mxcs = self._get_media_mxcs_in_room_txn(txn, room_id) + total_media_quarantined = 0 - next_token = self.get_current_events_token() + 1 + # Now update all the tables to set the quarantined_by flag - total_media_quarantined = 0 + txn.executemany(""" + UPDATE local_media_repository + SET quarantined_by = ? + WHERE media_id = ? + """, ((quarantined_by, media_id) for media_id in local_mxcs)) - while next_token: - sql = """ - SELECT stream_ordering, content FROM events - WHERE room_id = ? - AND stream_ordering < ? - AND contains_url = ? AND outlier = ? - ORDER BY stream_ordering DESC - LIMIT ? + txn.executemany( """ - txn.execute(sql, (room_id, next_token, True, False, 100)) - - next_token = None - local_media_mxcs = [] - remote_media_mxcs = [] - for stream_ordering, content_json in txn: - next_token = stream_ordering - content = json.loads(content_json) - - content_url = content.get("url") - thumbnail_url = content.get("info", {}).get("thumbnail_url") - - for url in (content_url, thumbnail_url): - if not url: - continue - matches = mxc_re.match(url) - if matches: - hostname = matches.group(1) - media_id = matches.group(2) - if hostname == self.hostname: - local_media_mxcs.append(media_id) - else: - remote_media_mxcs.append((hostname, media_id)) - - # Now update all the tables to set the quarantined_by flag - - txn.executemany(""" - UPDATE local_media_repository + UPDATE remote_media_cache SET quarantined_by = ? - WHERE media_id = ? - """, ((quarantined_by, media_id) for media_id in local_media_mxcs)) - - txn.executemany( - """ - UPDATE remote_media_cache - SET quarantined_by = ? - WHERE media_origin AND media_id = ? - """, - ( - (quarantined_by, origin, media_id) - for origin, media_id in remote_media_mxcs - ) + WHERE media_origin = ? AND media_id = ? + """, + ( + (quarantined_by, origin, media_id) + for origin, media_id in remote_mxcs ) + ) - total_media_quarantined += len(local_media_mxcs) - total_media_quarantined += len(remote_media_mxcs) + total_media_quarantined += len(local_mxcs) + total_media_quarantined += len(remote_mxcs) return total_media_quarantined - return self.runInteraction("get_media_ids_in_room", _get_media_ids_in_room) + return self.runInteraction( + "quarantine_media_in_room", + _quarantine_media_in_room_txn, + ) + + def _get_media_mxcs_in_room_txn(self, txn, room_id): + """Retrieves all the local and remote media MXC URIs in a given room + + Args: + txn (cursor) + room_id (str) + + Returns: + The local and remote media as a lists of tuples where the key is + the hostname and the value is the media ID. + """ + mxc_re = re.compile("^mxc://([^/]+)/([^/#?]+)") + + next_token = self.get_current_events_token() + 1 + local_media_mxcs = [] + remote_media_mxcs = [] + + while next_token: + sql = """ + SELECT stream_ordering, content FROM events + WHERE room_id = ? + AND stream_ordering < ? + AND contains_url = ? AND outlier = ? + ORDER BY stream_ordering DESC + LIMIT ? + """ + txn.execute(sql, (room_id, next_token, True, False, 100)) + + next_token = None + for stream_ordering, content_json in txn: + next_token = stream_ordering + content = json.loads(content_json) + + content_url = content.get("url") + thumbnail_url = content.get("info", {}).get("thumbnail_url") + + for url in (content_url, thumbnail_url): + if not url: + continue + matches = mxc_re.match(url) + if matches: + hostname = matches.group(1) + media_id = matches.group(2) + if hostname == self.hostname: + local_media_mxcs.append(media_id) + else: + remote_media_mxcs.append((hostname, media_id)) + + return local_media_mxcs, remote_media_mxcs diff --git a/synapse/storage/roommember.py b/synapse/storage/roommember.py index c1ca299285..d662d1cfc0 100644 --- a/synapse/storage/roommember.py +++ b/synapse/storage/roommember.py @@ -1,5 +1,6 @@ # -*- coding: utf-8 -*- # Copyright 2014-2016 OpenMarket Ltd +# Copyright 2018 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. @@ -17,7 +18,7 @@ from twisted.internet import defer from collections import namedtuple -from ._base import SQLBaseStore +from synapse.storage.events import EventsWorkerStore from synapse.util.async import Linearizer from synapse.util.caches import intern_string from synapse.util.caches.descriptors import cached, cachedInlineCallbacks @@ -37,6 +38,11 @@ RoomsForUser = namedtuple( ("room_id", "sender", "membership", "event_id", "stream_ordering") ) +GetRoomsForUserWithStreamOrdering = namedtuple( + "_GetRoomsForUserWithStreamOrdering", + ("room_id", "stream_ordering",) +) + # We store this using a namedtuple so that we save about 3x space over using a # dict. @@ -48,97 +54,7 @@ ProfileInfo = namedtuple( _MEMBERSHIP_PROFILE_UPDATE_NAME = "room_membership_profile_update" -class RoomMemberStore(SQLBaseStore): - def __init__(self, db_conn, hs): - super(RoomMemberStore, self).__init__(db_conn, hs) - self.register_background_update_handler( - _MEMBERSHIP_PROFILE_UPDATE_NAME, self._background_add_membership_profile - ) - - def _store_room_members_txn(self, txn, events, backfilled): - """Store a room member in the database. - """ - self._simple_insert_many_txn( - txn, - table="room_memberships", - values=[ - { - "event_id": event.event_id, - "user_id": event.state_key, - "sender": event.user_id, - "room_id": event.room_id, - "membership": event.membership, - "display_name": event.content.get("displayname", None), - "avatar_url": event.content.get("avatar_url", None), - } - for event in events - ] - ) - - for event in events: - txn.call_after( - self._membership_stream_cache.entity_has_changed, - event.state_key, event.internal_metadata.stream_ordering - ) - txn.call_after( - self.get_invited_rooms_for_user.invalidate, (event.state_key,) - ) - - # We update the local_invites table only if the event is "current", - # i.e., its something that has just happened. - # The only current event that can also be an outlier is if its an - # invite that has come in across federation. - is_new_state = not backfilled and ( - not event.internal_metadata.is_outlier() - or event.internal_metadata.is_invite_from_remote() - ) - is_mine = self.hs.is_mine_id(event.state_key) - if is_new_state and is_mine: - if event.membership == Membership.INVITE: - self._simple_insert_txn( - txn, - table="local_invites", - values={ - "event_id": event.event_id, - "invitee": event.state_key, - "inviter": event.sender, - "room_id": event.room_id, - "stream_id": event.internal_metadata.stream_ordering, - } - ) - else: - sql = ( - "UPDATE local_invites SET stream_id = ?, replaced_by = ? WHERE" - " room_id = ? AND invitee = ? AND locally_rejected is NULL" - " AND replaced_by is NULL" - ) - - txn.execute(sql, ( - event.internal_metadata.stream_ordering, - event.event_id, - event.room_id, - event.state_key, - )) - - @defer.inlineCallbacks - def locally_reject_invite(self, user_id, room_id): - sql = ( - "UPDATE local_invites SET stream_id = ?, locally_rejected = ? WHERE" - " room_id = ? AND invitee = ? AND locally_rejected is NULL" - " AND replaced_by is NULL" - ) - - def f(txn, stream_ordering): - txn.execute(sql, ( - stream_ordering, - True, - room_id, - user_id, - )) - - with self._stream_id_gen.get_next() as stream_ordering: - yield self.runInteraction("locally_reject_invite", f, stream_ordering) - +class RoomMemberWorkerStore(EventsWorkerStore): @cachedInlineCallbacks(max_entries=100000, iterable=True, cache_context=True) def get_hosts_in_room(self, room_id, cache_context): """Returns the set of all hosts currently in the room @@ -270,12 +186,32 @@ class RoomMemberStore(SQLBaseStore): return results @cachedInlineCallbacks(max_entries=500000, iterable=True) - def get_rooms_for_user(self, user_id): + def get_rooms_for_user_with_stream_ordering(self, user_id): """Returns a set of room_ids the user is currently joined to + + Args: + user_id (str) + + Returns: + Deferred[frozenset[GetRoomsForUserWithStreamOrdering]]: Returns + the rooms the user is in currently, along with the stream ordering + of the most recent join for that user and room. """ rooms = yield self.get_rooms_for_user_where_membership_is( user_id, membership_list=[Membership.JOIN], ) + defer.returnValue(frozenset( + GetRoomsForUserWithStreamOrdering(r.room_id, r.stream_ordering) + for r in rooms + )) + + @defer.inlineCallbacks + def get_rooms_for_user(self, user_id, on_invalidate=None): + """Returns a set of room_ids the user is currently joined to + """ + rooms = yield self.get_rooms_for_user_with_stream_ordering( + user_id, on_invalidate=on_invalidate, + ) defer.returnValue(frozenset(r.room_id for r in rooms)) @cachedInlineCallbacks(max_entries=500000, cache_context=True, iterable=True) @@ -295,89 +231,6 @@ class RoomMemberStore(SQLBaseStore): defer.returnValue(user_who_share_room) - def forget(self, user_id, room_id): - """Indicate that user_id wishes to discard history for room_id.""" - def f(txn): - sql = ( - "UPDATE" - " room_memberships" - " SET" - " forgotten = 1" - " WHERE" - " user_id = ?" - " AND" - " room_id = ?" - ) - txn.execute(sql, (user_id, room_id)) - - txn.call_after(self.was_forgotten_at.invalidate_all) - txn.call_after(self.did_forget.invalidate, (user_id, room_id)) - self._invalidate_cache_and_stream( - txn, self.who_forgot_in_room, (room_id,) - ) - return self.runInteraction("forget_membership", f) - - @cachedInlineCallbacks(num_args=2) - def did_forget(self, user_id, room_id): - """Returns whether user_id has elected to discard history for room_id. - - Returns False if they have since re-joined.""" - def f(txn): - sql = ( - "SELECT" - " COUNT(*)" - " FROM" - " room_memberships" - " WHERE" - " user_id = ?" - " AND" - " room_id = ?" - " AND" - " forgotten = 0" - ) - txn.execute(sql, (user_id, room_id)) - rows = txn.fetchall() - return rows[0][0] - count = yield self.runInteraction("did_forget_membership", f) - defer.returnValue(count == 0) - - @cachedInlineCallbacks(num_args=3) - def was_forgotten_at(self, user_id, room_id, event_id): - """Returns whether user_id has elected to discard history for room_id at - event_id. - - event_id must be a membership event.""" - def f(txn): - sql = ( - "SELECT" - " forgotten" - " FROM" - " room_memberships" - " WHERE" - " user_id = ?" - " AND" - " room_id = ?" - " AND" - " event_id = ?" - ) - txn.execute(sql, (user_id, room_id, event_id)) - rows = txn.fetchall() - return rows[0][0] - forgot = yield self.runInteraction("did_forget_membership_at", f) - defer.returnValue(forgot == 1) - - @cached() - def who_forgot_in_room(self, room_id): - return self._simple_select_list( - table="room_memberships", - retcols=("user_id", "event_id"), - keyvalues={ - "room_id": room_id, - "forgotten": 1, - }, - desc="who_forgot" - ) - def get_joined_users_from_context(self, event, context): state_group = context.state_group if not state_group: @@ -600,6 +453,185 @@ class RoomMemberStore(SQLBaseStore): defer.returnValue(joined_hosts) + @cached(max_entries=10000, iterable=True) + def _get_joined_hosts_cache(self, room_id): + return _JoinedHostsCache(self, room_id) + + @cached() + def who_forgot_in_room(self, room_id): + return self._simple_select_list( + table="room_memberships", + retcols=("user_id", "event_id"), + keyvalues={ + "room_id": room_id, + "forgotten": 1, + }, + desc="who_forgot" + ) + + +class RoomMemberStore(RoomMemberWorkerStore): + def __init__(self, db_conn, hs): + super(RoomMemberStore, self).__init__(db_conn, hs) + self.register_background_update_handler( + _MEMBERSHIP_PROFILE_UPDATE_NAME, self._background_add_membership_profile + ) + + def _store_room_members_txn(self, txn, events, backfilled): + """Store a room member in the database. + """ + self._simple_insert_many_txn( + txn, + table="room_memberships", + values=[ + { + "event_id": event.event_id, + "user_id": event.state_key, + "sender": event.user_id, + "room_id": event.room_id, + "membership": event.membership, + "display_name": event.content.get("displayname", None), + "avatar_url": event.content.get("avatar_url", None), + } + for event in events + ] + ) + + for event in events: + txn.call_after( + self._membership_stream_cache.entity_has_changed, + event.state_key, event.internal_metadata.stream_ordering + ) + txn.call_after( + self.get_invited_rooms_for_user.invalidate, (event.state_key,) + ) + + # We update the local_invites table only if the event is "current", + # i.e., its something that has just happened. + # The only current event that can also be an outlier is if its an + # invite that has come in across federation. + is_new_state = not backfilled and ( + not event.internal_metadata.is_outlier() + or event.internal_metadata.is_invite_from_remote() + ) + is_mine = self.hs.is_mine_id(event.state_key) + if is_new_state and is_mine: + if event.membership == Membership.INVITE: + self._simple_insert_txn( + txn, + table="local_invites", + values={ + "event_id": event.event_id, + "invitee": event.state_key, + "inviter": event.sender, + "room_id": event.room_id, + "stream_id": event.internal_metadata.stream_ordering, + } + ) + else: + sql = ( + "UPDATE local_invites SET stream_id = ?, replaced_by = ? WHERE" + " room_id = ? AND invitee = ? AND locally_rejected is NULL" + " AND replaced_by is NULL" + ) + + txn.execute(sql, ( + event.internal_metadata.stream_ordering, + event.event_id, + event.room_id, + event.state_key, + )) + + @defer.inlineCallbacks + def locally_reject_invite(self, user_id, room_id): + sql = ( + "UPDATE local_invites SET stream_id = ?, locally_rejected = ? WHERE" + " room_id = ? AND invitee = ? AND locally_rejected is NULL" + " AND replaced_by is NULL" + ) + + def f(txn, stream_ordering): + txn.execute(sql, ( + stream_ordering, + True, + room_id, + user_id, + )) + + with self._stream_id_gen.get_next() as stream_ordering: + yield self.runInteraction("locally_reject_invite", f, stream_ordering) + + def forget(self, user_id, room_id): + """Indicate that user_id wishes to discard history for room_id.""" + def f(txn): + sql = ( + "UPDATE" + " room_memberships" + " SET" + " forgotten = 1" + " WHERE" + " user_id = ?" + " AND" + " room_id = ?" + ) + txn.execute(sql, (user_id, room_id)) + + txn.call_after(self.was_forgotten_at.invalidate_all) + txn.call_after(self.did_forget.invalidate, (user_id, room_id)) + self._invalidate_cache_and_stream( + txn, self.who_forgot_in_room, (room_id,) + ) + return self.runInteraction("forget_membership", f) + + @cachedInlineCallbacks(num_args=2) + def did_forget(self, user_id, room_id): + """Returns whether user_id has elected to discard history for room_id. + + Returns False if they have since re-joined.""" + def f(txn): + sql = ( + "SELECT" + " COUNT(*)" + " FROM" + " room_memberships" + " WHERE" + " user_id = ?" + " AND" + " room_id = ?" + " AND" + " forgotten = 0" + ) + txn.execute(sql, (user_id, room_id)) + rows = txn.fetchall() + return rows[0][0] + count = yield self.runInteraction("did_forget_membership", f) + defer.returnValue(count == 0) + + @cachedInlineCallbacks(num_args=3) + def was_forgotten_at(self, user_id, room_id, event_id): + """Returns whether user_id has elected to discard history for room_id at + event_id. + + event_id must be a membership event.""" + def f(txn): + sql = ( + "SELECT" + " forgotten" + " FROM" + " room_memberships" + " WHERE" + " user_id = ?" + " AND" + " room_id = ?" + " AND" + " event_id = ?" + ) + txn.execute(sql, (user_id, room_id, event_id)) + rows = txn.fetchall() + return rows[0][0] + forgot = yield self.runInteraction("did_forget_membership_at", f) + defer.returnValue(forgot == 1) + @defer.inlineCallbacks def _background_add_membership_profile(self, progress, batch_size): target_min_stream_id = progress.get( @@ -675,10 +707,6 @@ class RoomMemberStore(SQLBaseStore): defer.returnValue(result) - @cached(max_entries=10000, iterable=True) - def _get_joined_hosts_cache(self, room_id): - return _JoinedHostsCache(self, room_id) - class _JoinedHostsCache(object): """Cache for joined hosts in a room that is optimised to handle updates diff --git a/synapse/storage/schema/delta/27/ts.py b/synapse/storage/schema/delta/27/ts.py index c0176c41ee..6df57b5206 100644 --- a/synapse/storage/schema/delta/27/ts.py +++ b/synapse/storage/schema/delta/27/ts.py @@ -45,7 +45,7 @@ def run_create(cur, database_engine, *args, **kwargs): "max_stream_id_exclusive": max_stream_id + 1, "rows_inserted": 0, } - progress_json = ujson.dumps(progress) + progress_json = simplejson.dumps(progress) sql = ( "INSERT into background_updates (update_name, progress_json)" diff --git a/synapse/storage/schema/delta/38/postgres_fts_gist.sql b/synapse/storage/schema/delta/38/postgres_fts_gist.sql index f090a7b75a..515e6b8e84 100644 --- a/synapse/storage/schema/delta/38/postgres_fts_gist.sql +++ b/synapse/storage/schema/delta/38/postgres_fts_gist.sql @@ -13,5 +13,7 @@ * limitations under the License. */ - INSERT into background_updates (update_name, progress_json) - VALUES ('event_search_postgres_gist', '{}'); +-- We no longer do this given we back it out again in schema 47 + +-- INSERT into background_updates (update_name, progress_json) +-- VALUES ('event_search_postgres_gist', '{}'); diff --git a/synapse/storage/schema/delta/47/last_access_media.sql b/synapse/storage/schema/delta/47/last_access_media.sql new file mode 100644 index 0000000000..f505fb22b5 --- /dev/null +++ b/synapse/storage/schema/delta/47/last_access_media.sql @@ -0,0 +1,16 @@ +/* Copyright 2018 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. + * 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. + */ + +ALTER TABLE local_media_repository ADD COLUMN last_access_ts BIGINT; diff --git a/synapse/storage/schema/delta/47/postgres_fts_gin.sql b/synapse/storage/schema/delta/47/postgres_fts_gin.sql new file mode 100644 index 0000000000..31d7a817eb --- /dev/null +++ b/synapse/storage/schema/delta/47/postgres_fts_gin.sql @@ -0,0 +1,17 @@ +/* Copyright 2018 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. + * 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_search_postgres_gin', '{}'); diff --git a/synapse/storage/schema/delta/47/push_actions_staging.sql b/synapse/storage/schema/delta/47/push_actions_staging.sql new file mode 100644 index 0000000000..edccf4a96f --- /dev/null +++ b/synapse/storage/schema/delta/47/push_actions_staging.sql @@ -0,0 +1,28 @@ +/* Copyright 2018 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. + * 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. + */ + +-- Temporary staging area for push actions that have been calculated for an +-- event, but the event hasn't yet been persisted. +-- When the event is persisted the rows are moved over to the +-- event_push_actions table. +CREATE TABLE event_push_actions_staging ( + event_id TEXT NOT NULL, + user_id TEXT NOT NULL, + actions TEXT NOT NULL, + notif SMALLINT NOT NULL, + highlight SMALLINT NOT NULL +); + +CREATE INDEX event_push_actions_staging_id ON event_push_actions_staging(event_id); diff --git a/synapse/storage/schema/delta/47/state_group_seq.py b/synapse/storage/schema/delta/47/state_group_seq.py new file mode 100644 index 0000000000..f6766501d2 --- /dev/null +++ b/synapse/storage/schema/delta/47/state_group_seq.py @@ -0,0 +1,37 @@ +# Copyright 2018 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. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +from synapse.storage.engines import PostgresEngine + + +def run_create(cur, database_engine, *args, **kwargs): + if isinstance(database_engine, PostgresEngine): + # if we already have some state groups, we want to start making new + # ones with a higher id. + cur.execute("SELECT max(id) FROM state_groups") + row = cur.fetchone() + + if row[0] is None: + start_val = 1 + else: + start_val = row[0] + 1 + + cur.execute( + "CREATE SEQUENCE state_group_id_seq START WITH %s", + (start_val, ), + ) + + +def run_upgrade(*args, **kwargs): + pass diff --git a/synapse/storage/search.py b/synapse/storage/search.py index c19e4ea449..984643b057 100644 --- a/synapse/storage/search.py +++ b/synapse/storage/search.py @@ -13,25 +13,32 @@ # See the License for the specific language governing permissions and # limitations under the License. +from collections import namedtuple +import logging +import re +import simplejson as json + from twisted.internet import defer from .background_updates import BackgroundUpdateStore from synapse.api.errors import SynapseError from synapse.storage.engines import PostgresEngine, Sqlite3Engine -import logging -import re -import simplejson as json - logger = logging.getLogger(__name__) +SearchEntry = namedtuple('SearchEntry', [ + 'key', 'value', 'event_id', 'room_id', 'stream_ordering', + 'origin_server_ts', +]) + class SearchStore(BackgroundUpdateStore): EVENT_SEARCH_UPDATE_NAME = "event_search" EVENT_SEARCH_ORDER_UPDATE_NAME = "event_search_order" EVENT_SEARCH_USE_GIST_POSTGRES_NAME = "event_search_postgres_gist" + EVENT_SEARCH_USE_GIN_POSTGRES_NAME = "event_search_postgres_gin" def __init__(self, db_conn, hs): super(SearchStore, self).__init__(db_conn, hs) @@ -42,23 +49,34 @@ class SearchStore(BackgroundUpdateStore): self.EVENT_SEARCH_ORDER_UPDATE_NAME, self._background_reindex_search_order ) - self.register_background_update_handler( + + # we used to have a background update to turn the GIN index into a + # GIST one; we no longer do that (obviously) because we actually want + # a GIN index. However, it's possible that some people might still have + # the background update queued, so we register a handler to clear the + # background update. + self.register_noop_background_update( self.EVENT_SEARCH_USE_GIST_POSTGRES_NAME, - self._background_reindex_gist_search + ) + + self.register_background_update_handler( + self.EVENT_SEARCH_USE_GIN_POSTGRES_NAME, + self._background_reindex_gin_search ) @defer.inlineCallbacks def _background_reindex_search(self, progress, batch_size): + # we work through the events table from highest stream id to lowest target_min_stream_id = progress["target_min_stream_id_inclusive"] max_stream_id = progress["max_stream_id_exclusive"] rows_inserted = progress.get("rows_inserted", 0) - INSERT_CLUMP_SIZE = 1000 TYPES = ["m.room.name", "m.room.message", "m.room.topic"] def reindex_search_txn(txn): sql = ( - "SELECT stream_ordering, event_id, room_id, type, content FROM events" + "SELECT stream_ordering, event_id, room_id, type, content, " + " origin_server_ts FROM events" " WHERE ? <= stream_ordering AND stream_ordering < ?" " AND (%s)" " ORDER BY stream_ordering DESC" @@ -67,6 +85,10 @@ class SearchStore(BackgroundUpdateStore): txn.execute(sql, (target_min_stream_id, max_stream_id, batch_size)) + # we could stream straight from the results into + # store_search_entries_txn with a generator function, but that + # would mean having two cursors open on the database at once. + # Instead we just build a list of results. rows = self.cursor_to_dict(txn) if not rows: return 0 @@ -79,6 +101,8 @@ class SearchStore(BackgroundUpdateStore): event_id = row["event_id"] room_id = row["room_id"] etype = row["type"] + stream_ordering = row["stream_ordering"] + origin_server_ts = row["origin_server_ts"] try: content = json.loads(row["content"]) except Exception: @@ -93,6 +117,8 @@ class SearchStore(BackgroundUpdateStore): elif etype == "m.room.name": key = "content.name" value = content["name"] + else: + raise Exception("unexpected event type %s" % etype) except (KeyError, AttributeError): # If the event is missing a necessary field then # skip over it. @@ -103,25 +129,16 @@ class SearchStore(BackgroundUpdateStore): # then skip over it continue - event_search_rows.append((event_id, room_id, key, value)) + event_search_rows.append(SearchEntry( + key=key, + value=value, + event_id=event_id, + room_id=room_id, + stream_ordering=stream_ordering, + origin_server_ts=origin_server_ts, + )) - if isinstance(self.database_engine, PostgresEngine): - sql = ( - "INSERT INTO event_search (event_id, room_id, key, vector)" - " VALUES (?,?,?,to_tsvector('english', ?))" - ) - elif isinstance(self.database_engine, Sqlite3Engine): - sql = ( - "INSERT INTO event_search (event_id, room_id, key, value)" - " VALUES (?,?,?,?)" - ) - else: - # This should be unreachable. - raise Exception("Unrecognized database engine") - - for index in range(0, len(event_search_rows), INSERT_CLUMP_SIZE): - clump = event_search_rows[index:index + INSERT_CLUMP_SIZE] - txn.executemany(sql, clump) + self.store_search_entries_txn(txn, event_search_rows) progress = { "target_min_stream_id_inclusive": target_min_stream_id, @@ -145,25 +162,48 @@ class SearchStore(BackgroundUpdateStore): defer.returnValue(result) @defer.inlineCallbacks - def _background_reindex_gist_search(self, progress, batch_size): + def _background_reindex_gin_search(self, progress, batch_size): + """This handles old synapses which used GIST indexes, if any; + converting them back to be GIN as per the actual schema. + """ + def create_index(conn): conn.rollback() - conn.set_session(autocommit=True) - c = conn.cursor() - c.execute( - "CREATE INDEX CONCURRENTLY event_search_fts_idx_gist" - " ON event_search USING GIST (vector)" - ) + # we have to set autocommit, because postgres refuses to + # CREATE INDEX CONCURRENTLY without it. + conn.set_session(autocommit=True) - c.execute("DROP INDEX event_search_fts_idx") + try: + c = conn.cursor() - conn.set_session(autocommit=False) + # if we skipped the conversion to GIST, we may already/still + # have an event_search_fts_idx; unfortunately postgres 9.4 + # doesn't support CREATE INDEX IF EXISTS so we just catch the + # exception and ignore it. + import psycopg2 + try: + c.execute( + "CREATE INDEX CONCURRENTLY event_search_fts_idx" + " ON event_search USING GIN (vector)" + ) + except psycopg2.ProgrammingError as e: + logger.warn( + "Ignoring error %r when trying to switch from GIST to GIN", + e + ) + + # we should now be able to delete the GIST index. + c.execute( + "DROP INDEX IF EXISTS event_search_fts_idx_gist" + ) + finally: + conn.set_session(autocommit=False) if isinstance(self.database_engine, PostgresEngine): yield self.runWithConnection(create_index) - yield self._end_background_update(self.EVENT_SEARCH_USE_GIST_POSTGRES_NAME) + yield self._end_background_update(self.EVENT_SEARCH_USE_GIN_POSTGRES_NAME) defer.returnValue(1) @defer.inlineCallbacks @@ -242,6 +282,85 @@ class SearchStore(BackgroundUpdateStore): defer.returnValue(num_rows) + def store_event_search_txn(self, txn, event, key, value): + """Add event to the search table + + Args: + txn (cursor): + event (EventBase): + key (str): + value (str): + """ + self.store_search_entries_txn( + txn, + (SearchEntry( + key=key, + value=value, + event_id=event.event_id, + room_id=event.room_id, + stream_ordering=event.internal_metadata.stream_ordering, + origin_server_ts=event.origin_server_ts, + ),), + ) + + def store_search_entries_txn(self, txn, entries): + """Add entries to the search table + + Args: + txn (cursor): + entries (iterable[SearchEntry]): + entries to be added to the table + """ + if isinstance(self.database_engine, PostgresEngine): + sql = ( + "INSERT INTO event_search" + " (event_id, room_id, key, vector, stream_ordering, origin_server_ts)" + " VALUES (?,?,?,to_tsvector('english', ?),?,?)" + ) + + args = (( + entry.event_id, entry.room_id, entry.key, entry.value, + entry.stream_ordering, entry.origin_server_ts, + ) for entry in entries) + + # inserts to a GIN index are normally batched up into a pending + # list, and then all committed together once the list gets to a + # certain size. The trouble with that is that postgres (pre-9.5) + # uses work_mem to determine the length of the list, and work_mem + # is typically very large. + # + # We therefore reduce work_mem while we do the insert. + # + # (postgres 9.5 uses the separate gin_pending_list_limit setting, + # so doesn't suffer the same problem, but changing work_mem will + # be harmless) + # + # Note that we don't need to worry about restoring it on + # exception, because exceptions will cause the transaction to be + # rolled back, including the effects of the SET command. + # + # Also: we use SET rather than SET LOCAL because there's lots of + # other stuff going on in this transaction, which want to have the + # normal work_mem setting. + + txn.execute("SET work_mem='256kB'") + txn.executemany(sql, args) + txn.execute("RESET work_mem") + + elif isinstance(self.database_engine, Sqlite3Engine): + sql = ( + "INSERT INTO event_search (event_id, room_id, key, value)" + " VALUES (?,?,?,?)" + ) + args = (( + entry.event_id, entry.room_id, entry.key, entry.value, + ) for entry in entries) + + txn.executemany(sql, args) + else: + # This should be unreachable. + raise Exception("Unrecognized database engine") + @defer.inlineCallbacks def search_msgs(self, room_ids, search_term, keys): """Performs a full text search over events with given keys. diff --git a/synapse/storage/signatures.py b/synapse/storage/signatures.py index 67d5d9969a..9e6eaaa532 100644 --- a/synapse/storage/signatures.py +++ b/synapse/storage/signatures.py @@ -22,12 +22,12 @@ from synapse.crypto.event_signing import compute_event_reference_hash from synapse.util.caches.descriptors import cached, cachedList -class SignatureStore(SQLBaseStore): - """Persistence for event signatures and hashes""" - +class SignatureWorkerStore(SQLBaseStore): @cached() def get_event_reference_hash(self, event_id): - return self._get_event_reference_hashes_txn(event_id) + # This is a dummy function to allow get_event_reference_hashes + # to use its cache + raise NotImplementedError() @cachedList(cached_method_name="get_event_reference_hash", list_name="event_ids", num_args=1) @@ -74,6 +74,10 @@ class SignatureStore(SQLBaseStore): txn.execute(query, (event_id, )) return {k: v for k, v in txn} + +class SignatureStore(SignatureWorkerStore): + """Persistence for event signatures and hashes""" + def _store_event_reference_hashes_txn(self, txn, events): """Store a hash for a PDU Args: diff --git a/synapse/storage/state.py b/synapse/storage/state.py index 360e3e4355..ffa4246031 100644 --- a/synapse/storage/state.py +++ b/synapse/storage/state.py @@ -42,11 +42,8 @@ class _GetStateGroupDelta(namedtuple("_GetStateGroupDelta", ("prev_group", "delt return len(self.delta_ids) if self.delta_ids else 0 -class StateGroupReadStore(SQLBaseStore): - """The read-only parts of StateGroupStore - - None of these functions write to the state tables, so are suitable for - including in the SlavedStores. +class StateGroupWorkerStore(SQLBaseStore): + """The parts of StateGroupStore that can be called from workers. """ STATE_GROUP_DEDUPLICATION_UPDATE_NAME = "state_group_state_deduplication" @@ -54,7 +51,7 @@ class StateGroupReadStore(SQLBaseStore): CURRENT_STATE_INDEX_UPDATE_NAME = "current_state_members_idx" def __init__(self, db_conn, hs): - super(StateGroupReadStore, self).__init__(db_conn, hs) + super(StateGroupWorkerStore, self).__init__(db_conn, hs) self._state_group_cache = DictionaryCache( "*stateGroupCache*", 100000 * CACHE_SIZE_FACTOR @@ -143,6 +140,20 @@ class StateGroupReadStore(SQLBaseStore): defer.returnValue(group_to_state) @defer.inlineCallbacks + def get_state_ids_for_group(self, state_group): + """Get the state IDs for the given state group + + Args: + state_group (int) + + Returns: + Deferred[dict]: Resolves to a map of (type, state_key) -> event_id + """ + group_to_state = yield self._get_state_for_groups((state_group,)) + + defer.returnValue(group_to_state[state_group]) + + @defer.inlineCallbacks def get_state_groups(self, room_id, event_ids): """ Get the state groups for the given list of event_ids @@ -229,6 +240,9 @@ class StateGroupReadStore(SQLBaseStore): ( "AND type = ? AND state_key = ?", (etype, state_key) + ) if state_key is not None else ( + "AND type = ?", + (etype,) ) for etype, state_key in types ] @@ -248,10 +262,19 @@ class StateGroupReadStore(SQLBaseStore): key = (typ, state_key) results[group][key] = event_id else: + where_args = [] + where_clauses = [] + wildcard_types = False if types is not None: - where_clause = "AND (%s)" % ( - " OR ".join(["(type = ? AND state_key = ?)"] * len(types)), - ) + for typ in types: + if typ[1] is None: + where_clauses.append("(type = ?)") + where_args.extend(typ[0]) + wildcard_types = True + else: + where_clauses.append("(type = ? AND state_key = ?)") + where_args.extend([typ[0], typ[1]]) + where_clause = "AND (%s)" % (" OR ".join(where_clauses)) else: where_clause = "" @@ -268,7 +291,7 @@ class StateGroupReadStore(SQLBaseStore): # after we finish deduping state, which requires this func) args = [next_group] if types: - args.extend(i for typ in types for i in typ) + args.extend(where_args) txn.execute( "SELECT type, state_key, event_id FROM state_groups_state" @@ -281,9 +304,17 @@ class StateGroupReadStore(SQLBaseStore): if (typ, state_key) not in results[group] ) - # If the lengths match then we must have all the types, - # so no need to go walk further down the tree. - if types is not None and len(results[group]) == len(types): + # If the number of entries in the (type,state_key)->event_id dict + # matches the number of (type,state_keys) types we were searching + # for, then we must have found them all, so no need to go walk + # further down the tree... UNLESS our types filter contained + # wildcards (i.e. Nones) in which case we have to do an exhaustive + # search + if ( + types is not None and + not wildcard_types and + len(results[group]) == len(types) + ): break next_group = self._simple_select_one_onecol_txn( @@ -549,116 +580,66 @@ class StateGroupReadStore(SQLBaseStore): defer.returnValue(results) + def store_state_group(self, event_id, room_id, prev_group, delta_ids, + current_state_ids): + """Store a new set of state, returning a newly assigned state group. -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 + Args: + event_id (str): The event ID for which the state was calculated + room_id (str) + prev_group (int|None): A previous state group for the room, optional. + delta_ids (dict|None): The delta between state at `prev_group` and + `current_state_ids`, if `prev_group` was given. Same format as + `current_state_ids`. + current_state_ids (dict): The state to store. Map of (type, state_key) + to event_id. - if context.current_state_ids is None: + Returns: + Deferred[int]: The state group ID + """ + def _store_state_group_txn(txn): + if 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 + raise Exception("current_state_ids cannot be None") - state_groups[event.event_id] = context.state_group - - if self._have_persisted_state_group_txn(txn, context.state_group): - continue + state_group = self.database_engine.get_next_state_group_id(txn) self._simple_insert_txn( txn, table="state_groups", values={ - "id": context.state_group, - "room_id": event.room_id, - "event_id": event.event_id, + "id": state_group, + "room_id": room_id, + "event_id": 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: + if prev_group: is_in_db = self._simple_select_one_onecol_txn( txn, table="state_groups", - keyvalues={"id": context.prev_group}, + keyvalues={"id": 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,) + % (prev_group,) ) potential_hops = self._count_state_group_hops_txn( - txn, context.prev_group + txn, prev_group ) - if context.prev_group and potential_hops < MAX_STATE_DELTA_HOPS: + if 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, + "state_group": state_group, + "prev_state_group": prev_group, }, ) @@ -667,13 +648,13 @@ class StateStore(StateGroupReadStore, BackgroundUpdateStore): table="state_groups_state", values=[ { - "state_group": context.state_group, - "room_id": event.room_id, + "state_group": state_group, + "room_id": room_id, "type": key[0], "state_key": key[1], "event_id": state_id, } - for key, state_id in context.delta_ids.iteritems() + for key, state_id in delta_ids.iteritems() ], ) else: @@ -682,13 +663,13 @@ class StateStore(StateGroupReadStore, BackgroundUpdateStore): table="state_groups_state", values=[ { - "state_group": context.state_group, - "room_id": event.room_id, + "state_group": state_group, + "room_id": room_id, "type": key[0], "state_key": key[1], "event_id": state_id, } - for key, state_id in context.current_state_ids.iteritems() + for key, state_id in current_state_ids.iteritems() ], ) @@ -699,28 +680,14 @@ class StateStore(StateGroupReadStore, BackgroundUpdateStore): txn.call_after( self._state_group_cache.update, self._state_group_cache.sequence, - key=context.state_group, - value=dict(context.current_state_ids), + key=state_group, + value=dict(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() - ], - ) + return state_group - 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 - ) + return self.runInteraction("store_state_group", _store_state_group_txn) def _count_state_group_hops_txn(self, txn, state_group): """Given a state group, count how many hops there are in the tree. @@ -763,8 +730,79 @@ class StateStore(StateGroupReadStore, BackgroundUpdateStore): return count - def get_next_state_group(self): - return self._state_groups_id_gen.get_next() + +class StateStore(StateGroupWorkerStore, 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 _store_event_state_mappings_txn(self, txn, events_and_contexts): + state_groups = {} + for event, context in events_and_contexts: + if event.internal_metadata.is_outlier(): + 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 + + 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 + ) @defer.inlineCallbacks def _background_deduplicate_state(self, progress, batch_size): diff --git a/synapse/storage/stream.py b/synapse/storage/stream.py index 52bdce5be2..2956c3b3e0 100644 --- a/synapse/storage/stream.py +++ b/synapse/storage/stream.py @@ -35,13 +35,16 @@ what sort order was used: from twisted.internet import defer -from ._base import SQLBaseStore +from synapse.storage._base import SQLBaseStore +from synapse.storage.events import EventsWorkerStore + from synapse.util.caches.descriptors import cached -from synapse.api.constants import EventTypes from synapse.types import RoomStreamToken +from synapse.util.caches.stream_change_cache import StreamChangeCache from synapse.util.logcontext import make_deferred_yieldable, preserve_fn from synapse.storage.engines import PostgresEngine, Sqlite3Engine +import abc import logging @@ -143,81 +146,41 @@ def filter_to_clause(event_filter): return " AND ".join(clauses), args -class StreamStore(SQLBaseStore): - @defer.inlineCallbacks - def get_appservice_room_stream(self, service, from_key, to_key, limit=0): - # NB this lives here instead of appservice.py so we can reuse the - # 'private' StreamToken class in this file. - if limit: - limit = max(limit, MAX_STREAM_SIZE) - else: - limit = MAX_STREAM_SIZE - - # From and to keys should be integers from ordering. - from_id = RoomStreamToken.parse_stream_token(from_key) - to_id = RoomStreamToken.parse_stream_token(to_key) - - if from_key == to_key: - defer.returnValue(([], to_key)) - return - - # select all the events between from/to with a sensible limit - sql = ( - "SELECT e.event_id, e.room_id, e.type, s.state_key, " - "e.stream_ordering FROM events AS e " - "LEFT JOIN state_events as s ON " - "e.event_id = s.event_id " - "WHERE e.stream_ordering > ? AND e.stream_ordering <= ? " - "ORDER BY stream_ordering ASC LIMIT %(limit)d " - ) % { - "limit": limit - } - - def f(txn): - # pull out all the events between the tokens - txn.execute(sql, (from_id.stream, to_id.stream,)) - rows = self.cursor_to_dict(txn) - - # Logic: - # - We want ALL events which match the AS room_id regex - # - We want ALL events which match the rooms represented by the AS - # room_alias regex - # - We want ALL events for rooms that AS users have joined. - # This is currently supported via get_app_service_rooms (which is - # used for the Notifier listener rooms). We can't reasonably make a - # SQL query for these room IDs, so we'll pull all the events between - # from/to and filter in python. - rooms_for_as = self._get_app_service_rooms_txn(txn, service) - room_ids_for_as = [r.room_id for r in rooms_for_as] - - def app_service_interested(row): - if row["room_id"] in room_ids_for_as: - return True - - if row["type"] == EventTypes.Member: - if service.is_interested_in_user(row.get("state_key")): - return True - return False +class StreamWorkerStore(EventsWorkerStore, SQLBaseStore): + """This is an abstract base class where subclasses must implement + `get_room_max_stream_ordering` and `get_room_min_stream_ordering` + which can be called in the initializer. + """ - return [r for r in rows if app_service_interested(r)] + __metaclass__ = abc.ABCMeta - rows = yield self.runInteraction("get_appservice_room_stream", f) + def __init__(self, db_conn, hs): + super(StreamWorkerStore, self).__init__(db_conn, hs) - ret = yield self._get_events( - [r["event_id"] for r in rows], - get_prev_content=True + events_max = self.get_room_max_stream_ordering() + event_cache_prefill, min_event_val = self._get_cache_dict( + db_conn, "events", + entity_column="room_id", + stream_column="stream_ordering", + max_value=events_max, + ) + self._events_stream_cache = StreamChangeCache( + "EventsRoomStreamChangeCache", min_event_val, + prefilled_cache=event_cache_prefill, + ) + self._membership_stream_cache = StreamChangeCache( + "MembershipStreamChangeCache", events_max, ) - self._set_before_and_after(ret, rows, topo_order=from_id is None) + self._stream_order_on_start = self.get_room_max_stream_ordering() - if rows: - key = "s%d" % max(r["stream_ordering"] for r in rows) - else: - # Assume we didn't get anything because there was nothing to - # get. - key = to_key + @abc.abstractmethod + def get_room_max_stream_ordering(self): + raise NotImplementedError() - defer.returnValue((ret, key)) + @abc.abstractmethod + def get_room_min_stream_ordering(self): + raise NotImplementedError() @defer.inlineCallbacks def get_room_events_stream_for_rooms(self, room_ids, from_key, to_key, limit=0, @@ -381,88 +344,6 @@ class StreamStore(SQLBaseStore): defer.returnValue(ret) @defer.inlineCallbacks - def paginate_room_events(self, room_id, from_key, to_key=None, - direction='b', limit=-1, event_filter=None): - # Tokens really represent positions between elements, but we use - # the convention of pointing to the event before the gap. Hence - # we have a bit of asymmetry when it comes to equalities. - args = [False, room_id] - if direction == 'b': - order = "DESC" - bounds = upper_bound( - RoomStreamToken.parse(from_key), self.database_engine - ) - if to_key: - bounds = "%s AND %s" % (bounds, lower_bound( - RoomStreamToken.parse(to_key), self.database_engine - )) - else: - order = "ASC" - bounds = lower_bound( - RoomStreamToken.parse(from_key), self.database_engine - ) - if to_key: - bounds = "%s AND %s" % (bounds, upper_bound( - RoomStreamToken.parse(to_key), self.database_engine - )) - - filter_clause, filter_args = filter_to_clause(event_filter) - - if filter_clause: - bounds += " AND " + filter_clause - args.extend(filter_args) - - if int(limit) > 0: - args.append(int(limit)) - limit_str = " LIMIT ?" - else: - limit_str = "" - - sql = ( - "SELECT * FROM events" - " WHERE outlier = ? AND room_id = ? AND %(bounds)s" - " ORDER BY topological_ordering %(order)s," - " stream_ordering %(order)s %(limit)s" - ) % { - "bounds": bounds, - "order": order, - "limit": limit_str - } - - def f(txn): - txn.execute(sql, args) - - rows = self.cursor_to_dict(txn) - - if rows: - topo = rows[-1]["topological_ordering"] - toke = rows[-1]["stream_ordering"] - if direction == 'b': - # Tokens are positions between events. - # This token points *after* the last event in the chunk. - # We need it to point to the event before it in the chunk - # when we are going backwards so we subtract one from the - # stream part. - toke -= 1 - next_token = str(RoomStreamToken(topo, toke)) - else: - # TODO (erikj): We should work out what to do here instead. - next_token = to_key if to_key else from_key - - return rows, next_token, - - rows, token = yield self.runInteraction("paginate_room_events", f) - - events = yield self._get_events( - [r["event_id"] for r in rows], - get_prev_content=True - ) - - self._set_before_and_after(events, rows) - - defer.returnValue((events, token)) - - @defer.inlineCallbacks def get_recent_events_for_room(self, room_id, limit, end_token, from_token=None): rows, token = yield self.get_recent_event_ids_for_room( room_id, limit, end_token, from_token @@ -534,6 +415,33 @@ class StreamStore(SQLBaseStore): "get_recent_events_for_room", get_recent_events_for_room_txn ) + def get_room_event_after_stream_ordering(self, room_id, stream_ordering): + """Gets details of the first event in a room at or after a stream ordering + + Args: + room_id (str): + stream_ordering (int): + + Returns: + Deferred[(int, int, str)]: + (stream ordering, topological ordering, event_id) + """ + def _f(txn): + sql = ( + "SELECT stream_ordering, topological_ordering, event_id" + " FROM events" + " WHERE room_id = ? AND stream_ordering >= ?" + " AND NOT outlier" + " ORDER BY stream_ordering" + " LIMIT 1" + ) + txn.execute(sql, (room_id, stream_ordering, )) + return txn.fetchone() + + return self.runInteraction( + "get_room_event_after_stream_ordering", _f, + ) + @defer.inlineCallbacks def get_room_events_max_id(self, room_id=None): """Returns the current token for rooms stream. @@ -542,7 +450,7 @@ class StreamStore(SQLBaseStore): `room_id` causes it to return the current room specific topological token. """ - token = yield self._stream_id_gen.get_current_token() + token = yield self.get_room_max_stream_ordering() if room_id is None: defer.returnValue("s%d" % (token,)) else: @@ -552,12 +460,6 @@ class StreamStore(SQLBaseStore): ) defer.returnValue("t%d-%d" % (topo, token)) - def get_room_max_stream_ordering(self): - return self._stream_id_gen.get_current_token() - - def get_room_min_stream_ordering(self): - return self._backfill_id_gen.get_current_token() - def get_stream_token_for_event(self, event_id): """The stream token for an event Args: @@ -832,3 +734,93 @@ class StreamStore(SQLBaseStore): def has_room_changed_since(self, room_id, stream_id): return self._events_stream_cache.has_entity_changed(room_id, stream_id) + + +class StreamStore(StreamWorkerStore): + def get_room_max_stream_ordering(self): + return self._stream_id_gen.get_current_token() + + def get_room_min_stream_ordering(self): + return self._backfill_id_gen.get_current_token() + + @defer.inlineCallbacks + def paginate_room_events(self, room_id, from_key, to_key=None, + direction='b', limit=-1, event_filter=None): + # Tokens really represent positions between elements, but we use + # the convention of pointing to the event before the gap. Hence + # we have a bit of asymmetry when it comes to equalities. + args = [False, room_id] + if direction == 'b': + order = "DESC" + bounds = upper_bound( + RoomStreamToken.parse(from_key), self.database_engine + ) + if to_key: + bounds = "%s AND %s" % (bounds, lower_bound( + RoomStreamToken.parse(to_key), self.database_engine + )) + else: + order = "ASC" + bounds = lower_bound( + RoomStreamToken.parse(from_key), self.database_engine + ) + if to_key: + bounds = "%s AND %s" % (bounds, upper_bound( + RoomStreamToken.parse(to_key), self.database_engine + )) + + filter_clause, filter_args = filter_to_clause(event_filter) + + if filter_clause: + bounds += " AND " + filter_clause + args.extend(filter_args) + + if int(limit) > 0: + args.append(int(limit)) + limit_str = " LIMIT ?" + else: + limit_str = "" + + sql = ( + "SELECT * FROM events" + " WHERE outlier = ? AND room_id = ? AND %(bounds)s" + " ORDER BY topological_ordering %(order)s," + " stream_ordering %(order)s %(limit)s" + ) % { + "bounds": bounds, + "order": order, + "limit": limit_str + } + + def f(txn): + txn.execute(sql, args) + + rows = self.cursor_to_dict(txn) + + if rows: + topo = rows[-1]["topological_ordering"] + toke = rows[-1]["stream_ordering"] + if direction == 'b': + # Tokens are positions between events. + # This token points *after* the last event in the chunk. + # We need it to point to the event before it in the chunk + # when we are going backwards so we subtract one from the + # stream part. + toke -= 1 + next_token = str(RoomStreamToken(topo, toke)) + else: + # TODO (erikj): We should work out what to do here instead. + next_token = to_key if to_key else from_key + + return rows, next_token, + + rows, token = yield self.runInteraction("paginate_room_events", f) + + events = yield self._get_events( + [r["event_id"] for r in rows], + get_prev_content=True + ) + + self._set_before_and_after(events, rows) + + defer.returnValue((events, token)) diff --git a/synapse/storage/tags.py b/synapse/storage/tags.py index 982a500520..13bff9f055 100644 --- a/synapse/storage/tags.py +++ b/synapse/storage/tags.py @@ -1,5 +1,6 @@ # -*- coding: utf-8 -*- # Copyright 2014-2016 OpenMarket Ltd +# Copyright 2018 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. @@ -13,7 +14,8 @@ # See the License for the specific language governing permissions and # limitations under the License. -from ._base import SQLBaseStore +from synapse.storage.account_data import AccountDataWorkerStore + from synapse.util.caches.descriptors import cached from twisted.internet import defer @@ -23,15 +25,7 @@ import logging logger = logging.getLogger(__name__) -class TagsStore(SQLBaseStore): - def get_max_account_data_stream_id(self): - """Get the current max stream id for the private user data stream - - Returns: - A deferred int. - """ - return self._account_data_id_gen.get_current_token() - +class TagsWorkerStore(AccountDataWorkerStore): @cached() def get_tags_for_user(self, user_id): """Get all the tags for a user. @@ -170,6 +164,8 @@ class TagsStore(SQLBaseStore): row["tag"]: json.loads(row["content"]) for row in rows }) + +class TagsStore(TagsWorkerStore): @defer.inlineCallbacks def add_tag_to_room(self, user_id, room_id, tag, content): """Add a tag to a room for a user. diff --git a/synapse/storage/user_directory.py b/synapse/storage/user_directory.py index c9bff408ef..dfdcbb3181 100644 --- a/synapse/storage/user_directory.py +++ b/synapse/storage/user_directory.py @@ -641,8 +641,12 @@ class UserDirectoryStore(SQLBaseStore): """ if self.hs.config.user_directory_search_all_users: - join_clause = "" - where_clause = "?<>''" # naughty hack to keep the same number of binds + # make s.user_id null to keep the ordering algorithm happy + join_clause = """ + CROSS JOIN (SELECT NULL as user_id) AS s + """ + join_args = () + where_clause = "1=1" else: join_clause = """ LEFT JOIN users_in_public_rooms AS p USING (user_id) @@ -651,6 +655,7 @@ class UserDirectoryStore(SQLBaseStore): WHERE user_id = ? AND share_private ) AS s USING (user_id) """ + join_args = (user_id,) where_clause = "(s.user_id IS NOT NULL OR p.user_id IS NOT NULL)" if isinstance(self.database_engine, PostgresEngine): @@ -692,7 +697,7 @@ class UserDirectoryStore(SQLBaseStore): avatar_url IS NULL LIMIT ? """ % (join_clause, where_clause) - args = (user_id, full_query, exact_query, prefix_query, limit + 1,) + args = join_args + (full_query, exact_query, prefix_query, limit + 1,) elif isinstance(self.database_engine, Sqlite3Engine): search_query = _parse_query_sqlite(search_term) @@ -710,7 +715,7 @@ class UserDirectoryStore(SQLBaseStore): avatar_url IS NULL LIMIT ? """ % (join_clause, where_clause) - args = (user_id, search_query, limit + 1) + args = join_args + (search_query, limit + 1) else: # This should be unreachable. raise Exception("Unrecognized database engine") diff --git a/synapse/types.py b/synapse/types.py index 6e76c016d9..7cb24cecb2 100644 --- a/synapse/types.py +++ b/synapse/types.py @@ -19,20 +19,59 @@ from synapse.api.errors import SynapseError from collections import namedtuple -Requester = namedtuple("Requester", [ +class Requester(namedtuple("Requester", [ "user", "access_token_id", "is_guest", "device_id", "app_service", -]) -""" -Represents the user making a request - -Attributes: - user (UserID): id of the user making the request - access_token_id (int|None): *ID* of the access token used for this - request, or None if it came via the appservice API or similar - is_guest (bool): True if the user making this request is a guest user - device_id (str|None): device_id which was set at authentication time - app_service (ApplicationService|None): the AS requesting on behalf of the user -""" +])): + """ + Represents the user making a request + + Attributes: + user (UserID): id of the user making the request + access_token_id (int|None): *ID* of the access token used for this + request, or None if it came via the appservice API or similar + is_guest (bool): True if the user making this request is a guest user + device_id (str|None): device_id which was set at authentication time + app_service (ApplicationService|None): the AS requesting on behalf of the user + """ + + def serialize(self): + """Converts self to a type that can be serialized as JSON, and then + deserialized by `deserialize` + + Returns: + dict + """ + return { + "user_id": self.user.to_string(), + "access_token_id": self.access_token_id, + "is_guest": self.is_guest, + "device_id": self.device_id, + "app_server_id": self.app_service.id if self.app_service else None, + } + + @staticmethod + def deserialize(store, input): + """Converts a dict that was produced by `serialize` back into a + Requester. + + Args: + store (DataStore): Used to convert AS ID to AS object + input (dict): A dict produced by `serialize` + + Returns: + Requester + """ + appservice = None + if input["app_server_id"]: + appservice = store.get_app_service_by_id(input["app_server_id"]) + + return Requester( + user=UserID.from_string(input["user_id"]), + access_token_id=input["access_token_id"], + is_guest=input["is_guest"], + device_id=input["device_id"], + app_service=appservice, + ) def create_requester(user_id, access_token_id=None, is_guest=False, diff --git a/synapse/util/caches/descriptors.py b/synapse/util/caches/descriptors.py index af65bfe7b8..bf3a66eae4 100644 --- a/synapse/util/caches/descriptors.py +++ b/synapse/util/caches/descriptors.py @@ -75,6 +75,7 @@ class Cache(object): self.cache = LruCache( max_size=max_entries, keylen=keylen, cache_type=cache_type, size_callback=(lambda d: len(d)) if iterable else None, + evicted_callback=self._on_evicted, ) self.name = name @@ -83,6 +84,9 @@ class Cache(object): self.thread = None self.metrics = register_cache(name, self.cache) + def _on_evicted(self, evicted_count): + self.metrics.inc_evictions(evicted_count) + def check_thread(self): expected_thread = self.thread if expected_thread is None: diff --git a/synapse/util/caches/dictionary_cache.py b/synapse/util/caches/dictionary_cache.py index d4105822b3..1709e8b429 100644 --- a/synapse/util/caches/dictionary_cache.py +++ b/synapse/util/caches/dictionary_cache.py @@ -132,9 +132,13 @@ class DictionaryCache(object): self._update_or_insert(key, value, known_absent) def _update_or_insert(self, key, value, known_absent): - entry = self.cache.setdefault(key, DictionaryEntry(False, set(), {})) + # We pop and reinsert as we need to tell the cache the size may have + # changed + + entry = self.cache.pop(key, DictionaryEntry(False, set(), {})) entry.value.update(value) entry.known_absent.update(known_absent) + self.cache[key] = entry def _insert(self, key, value, known_absent): self.cache[key] = DictionaryEntry(True, known_absent, value) diff --git a/synapse/util/caches/expiringcache.py b/synapse/util/caches/expiringcache.py index 6ad53a6390..0aa103eecb 100644 --- a/synapse/util/caches/expiringcache.py +++ b/synapse/util/caches/expiringcache.py @@ -79,7 +79,11 @@ class ExpiringCache(object): while self._max_len and len(self) > self._max_len: _key, value = self._cache.popitem(last=False) if self.iterable: - self._size_estimate -= len(value.value) + removed_len = len(value.value) + self.metrics.inc_evictions(removed_len) + self._size_estimate -= removed_len + else: + self.metrics.inc_evictions() def __getitem__(self, key): try: diff --git a/synapse/util/caches/lrucache.py b/synapse/util/caches/lrucache.py index cf5fbb679c..1c5a982094 100644 --- a/synapse/util/caches/lrucache.py +++ b/synapse/util/caches/lrucache.py @@ -49,7 +49,24 @@ class LruCache(object): Can also set callbacks on objects when getting/setting which are fired when that key gets invalidated/evicted. """ - def __init__(self, max_size, keylen=1, cache_type=dict, size_callback=None): + def __init__(self, max_size, keylen=1, cache_type=dict, size_callback=None, + evicted_callback=None): + """ + Args: + max_size (int): + + keylen (int): + + cache_type (type): + type of underlying cache to be used. Typically one of dict + or TreeCache. + + size_callback (func(V) -> int | None): + + evicted_callback (func(int)|None): + if not None, called on eviction with the size of the evicted + entry + """ cache = cache_type() self.cache = cache # Used for introspection. list_root = _Node(None, None, None, None) @@ -61,8 +78,10 @@ class LruCache(object): def evict(): while cache_len() > max_size: todelete = list_root.prev_node - delete_node(todelete) + evicted_len = delete_node(todelete) cache.pop(todelete.key, None) + if evicted_callback: + evicted_callback(evicted_len) def synchronized(f): @wraps(f) @@ -111,12 +130,15 @@ class LruCache(object): prev_node.next_node = next_node next_node.prev_node = prev_node + deleted_len = 1 if size_callback: - cached_cache_len[0] -= size_callback(node.value) + deleted_len = size_callback(node.value) + cached_cache_len[0] -= deleted_len for cb in node.callbacks: cb() node.callbacks.clear() + return deleted_len @synchronized def cache_get(key, default=None, callbacks=[]): @@ -132,14 +154,21 @@ class LruCache(object): def cache_set(key, value, callbacks=[]): node = cache.get(key, None) if node is not None: - if value != node.value: + # We sometimes store large objects, e.g. dicts, which cause + # the inequality check to take a long time. So let's only do + # the check if we have some callbacks to call. + if node.callbacks and value != node.value: for cb in node.callbacks: cb() node.callbacks.clear() - if size_callback: - cached_cache_len[0] -= size_callback(node.value) - cached_cache_len[0] += size_callback(value) + # We don't bother to protect this by value != node.value as + # generally size_callback will be cheap compared with equality + # checks. (For example, taking the size of two dicts is quicker + # than comparing them for equality.) + if size_callback: + cached_cache_len[0] -= size_callback(node.value) + cached_cache_len[0] += size_callback(value) node.callbacks.update(callbacks) diff --git a/synapse/util/file_consumer.py b/synapse/util/file_consumer.py new file mode 100644 index 0000000000..90a2608d6f --- /dev/null +++ b/synapse/util/file_consumer.py @@ -0,0 +1,139 @@ +# -*- coding: utf-8 -*- +# Copyright 2018 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. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +from twisted.internet import threads, reactor + +from synapse.util.logcontext import make_deferred_yieldable, preserve_fn + +import Queue + + +class BackgroundFileConsumer(object): + """A consumer that writes to a file like object. Supports both push + and pull producers + + Args: + file_obj (file): The file like object to write to. Closed when + finished. + """ + + # For PushProducers pause if we have this many unwritten slices + _PAUSE_ON_QUEUE_SIZE = 5 + # And resume once the size of the queue is less than this + _RESUME_ON_QUEUE_SIZE = 2 + + def __init__(self, file_obj): + self._file_obj = file_obj + + # Producer we're registered with + self._producer = None + + # True if PushProducer, false if PullProducer + self.streaming = False + + # For PushProducers, indicates whether we've paused the producer and + # need to call resumeProducing before we get more data. + self._paused_producer = False + + # Queue of slices of bytes to be written. When producer calls + # unregister a final None is sent. + self._bytes_queue = Queue.Queue() + + # Deferred that is resolved when finished writing + self._finished_deferred = None + + # If the _writer thread throws an exception it gets stored here. + self._write_exception = None + + def registerProducer(self, producer, streaming): + """Part of IConsumer interface + + Args: + producer (IProducer) + streaming (bool): True if push based producer, False if pull + based. + """ + if self._producer: + raise Exception("registerProducer called twice") + + self._producer = producer + self.streaming = streaming + self._finished_deferred = preserve_fn(threads.deferToThread)(self._writer) + if not streaming: + self._producer.resumeProducing() + + def unregisterProducer(self): + """Part of IProducer interface + """ + self._producer = None + if not self._finished_deferred.called: + self._bytes_queue.put_nowait(None) + + def write(self, bytes): + """Part of IProducer interface + """ + if self._write_exception: + raise self._write_exception + + if self._finished_deferred.called: + raise Exception("consumer has closed") + + self._bytes_queue.put_nowait(bytes) + + # If this is a PushProducer and the queue is getting behind + # then we pause the producer. + if self.streaming and self._bytes_queue.qsize() >= self._PAUSE_ON_QUEUE_SIZE: + self._paused_producer = True + self._producer.pauseProducing() + + def _writer(self): + """This is run in a background thread to write to the file. + """ + try: + while self._producer or not self._bytes_queue.empty(): + # If we've paused the producer check if we should resume the + # producer. + if self._producer and self._paused_producer: + if self._bytes_queue.qsize() <= self._RESUME_ON_QUEUE_SIZE: + reactor.callFromThread(self._resume_paused_producer) + + bytes = self._bytes_queue.get() + + # If we get a None (or empty list) then that's a signal used + # to indicate we should check if we should stop. + if bytes: + self._file_obj.write(bytes) + + # If its a pull producer then we need to explicitly ask for + # more stuff. + if not self.streaming and self._producer: + reactor.callFromThread(self._producer.resumeProducing) + except Exception as e: + self._write_exception = e + raise + finally: + self._file_obj.close() + + def wait(self): + """Returns a deferred that resolves when finished writing to file + """ + return make_deferred_yieldable(self._finished_deferred) + + def _resume_paused_producer(self): + """Gets called if we should resume producing after being paused + """ + if self._paused_producer and self._producer: + self._paused_producer = False + self._producer.resumeProducing() diff --git a/synapse/util/logcontext.py b/synapse/util/logcontext.py index 48c9f6802d..d660ec785b 100644 --- a/synapse/util/logcontext.py +++ b/synapse/util/logcontext.py @@ -52,13 +52,17 @@ except Exception: class LoggingContext(object): """Additional context for log formatting. Contexts are scoped within a "with" block. + Args: name (str): Name for the context for debugging. """ __slots__ = [ - "previous_context", "name", "usage_start", "usage_end", "main_thread", - "__dict__", "tag", "alive", + "previous_context", "name", "ru_stime", "ru_utime", + "db_txn_count", "db_txn_duration_ms", "db_sched_duration_ms", + "usage_start", "usage_end", + "main_thread", "alive", + "request", "tag", ] thread_local = threading.local() @@ -83,6 +87,9 @@ class LoggingContext(object): def add_database_transaction(self, duration_ms): pass + def add_database_scheduled(self, sched_ms): + pass + def __nonzero__(self): return False @@ -94,9 +101,17 @@ class LoggingContext(object): self.ru_stime = 0. self.ru_utime = 0. self.db_txn_count = 0 - self.db_txn_duration = 0. + + # ms spent waiting for db txns, excluding scheduling time + self.db_txn_duration_ms = 0 + + # ms spent waiting for db txns to be scheduled + self.db_sched_duration_ms = 0 + self.usage_start = None + self.usage_end = None self.main_thread = threading.current_thread() + self.request = None self.tag = "" self.alive = True @@ -105,7 +120,11 @@ class LoggingContext(object): @classmethod def current_context(cls): - """Get the current logging context from thread local storage""" + """Get the current logging context from thread local storage + + Returns: + LoggingContext: the current logging context + """ return getattr(cls.thread_local, "current_context", cls.sentinel) @classmethod @@ -155,11 +174,13 @@ class LoggingContext(object): self.alive = False def copy_to(self, record): - """Copy fields from this context to the record""" - for key, value in self.__dict__.items(): - setattr(record, key, value) + """Copy logging fields from this context to a log record or + another LoggingContext + """ - record.ru_utime, record.ru_stime = self.get_resource_usage() + # 'request' is the only field we currently use in the logger, so that's + # all we need to copy + record.request = self.request def start(self): if threading.current_thread() is not self.main_thread: @@ -194,7 +215,16 @@ class LoggingContext(object): def add_database_transaction(self, duration_ms): self.db_txn_count += 1 - self.db_txn_duration += duration_ms / 1000. + self.db_txn_duration_ms += duration_ms + + def add_database_scheduled(self, sched_ms): + """Record a use of the database pool + + Args: + sched_ms (int): number of milliseconds it took us to get a + connection + """ + self.db_sched_duration_ms += sched_ms class LoggingContextFilter(logging.Filter): @@ -262,43 +292,43 @@ class PreserveLoggingContext(object): def preserve_fn(f): - """Wraps a function, to ensure that the current context is restored after + """Function decorator which wraps the function with run_in_background""" + def g(*args, **kwargs): + return run_in_background(f, *args, **kwargs) + return g + + +def run_in_background(f, *args, **kwargs): + """Calls a function, ensuring that the current context is restored after return from the function, and that the sentinel context is set once the deferred returned by the funtion completes. Useful for wrapping functions that return a deferred which you don't yield on. """ - def reset_context(result): - LoggingContext.set_current_context(LoggingContext.sentinel) - return result - - def g(*args, **kwargs): - current = LoggingContext.current_context() - res = f(*args, **kwargs) - if isinstance(res, defer.Deferred) and not res.called: - # The function will have reset the context before returning, so - # we need to restore it now. - LoggingContext.set_current_context(current) - - # The original context will be restored when the deferred - # completes, but there is nothing waiting for it, so it will - # get leaked into the reactor or some other function which - # wasn't expecting it. We therefore need to reset the context - # here. - # - # (If this feels asymmetric, consider it this way: we are - # effectively forking a new thread of execution. We are - # probably currently within a ``with LoggingContext()`` block, - # which is supposed to have a single entry and exit point. But - # by spawning off another deferred, we are effectively - # adding a new exit point.) - res.addBoth(reset_context) - return res - return g + current = LoggingContext.current_context() + res = f(*args, **kwargs) + if isinstance(res, defer.Deferred) and not res.called: + # The function will have reset the context before returning, so + # we need to restore it now. + LoggingContext.set_current_context(current) + + # The original context will be restored when the deferred + # completes, but there is nothing waiting for it, so it will + # get leaked into the reactor or some other function which + # wasn't expecting it. We therefore need to reset the context + # here. + # + # (If this feels asymmetric, consider it this way: we are + # effectively forking a new thread of execution. We are + # probably currently within a ``with LoggingContext()`` block, + # which is supposed to have a single entry and exit point. But + # by spawning off another deferred, we are effectively + # adding a new exit point.) + res.addBoth(_set_context_cb, LoggingContext.sentinel) + return res -@defer.inlineCallbacks def make_deferred_yieldable(deferred): """Given a deferred, make it follow the Synapse logcontext rules: @@ -312,9 +342,16 @@ def make_deferred_yieldable(deferred): (This is more-or-less the opposite operation to preserve_fn.) """ - with PreserveLoggingContext(): - r = yield deferred - defer.returnValue(r) + if isinstance(deferred, defer.Deferred) and not deferred.called: + prev_context = LoggingContext.set_current_context(LoggingContext.sentinel) + deferred.addBoth(_set_context_cb, prev_context) + return deferred + + +def _set_context_cb(result, context): + """A callback function which just sets the logging context""" + LoggingContext.set_current_context(context) + return result # modules to ignore in `logcontext_tracer` diff --git a/synapse/util/metrics.py b/synapse/util/metrics.py index 4ea930d3e8..e4b5687a4b 100644 --- a/synapse/util/metrics.py +++ b/synapse/util/metrics.py @@ -27,25 +27,62 @@ logger = logging.getLogger(__name__) metrics = synapse.metrics.get_metrics_for(__name__) -block_timer = metrics.register_distribution( - "block_timer", - labels=["block_name"] +# total number of times we have hit this block +block_counter = metrics.register_counter( + "block_count", + labels=["block_name"], + alternative_names=( + # the following are all deprecated aliases for the same metric + metrics.name_prefix + x for x in ( + "_block_timer:count", + "_block_ru_utime:count", + "_block_ru_stime:count", + "_block_db_txn_count:count", + "_block_db_txn_duration:count", + ) + ) +) + +block_timer = metrics.register_counter( + "block_time_seconds", + labels=["block_name"], + alternative_names=( + metrics.name_prefix + "_block_timer:total", + ), ) -block_ru_utime = metrics.register_distribution( - "block_ru_utime", labels=["block_name"] +block_ru_utime = metrics.register_counter( + "block_ru_utime_seconds", labels=["block_name"], + alternative_names=( + metrics.name_prefix + "_block_ru_utime:total", + ), ) -block_ru_stime = metrics.register_distribution( - "block_ru_stime", labels=["block_name"] +block_ru_stime = metrics.register_counter( + "block_ru_stime_seconds", labels=["block_name"], + alternative_names=( + metrics.name_prefix + "_block_ru_stime:total", + ), ) -block_db_txn_count = metrics.register_distribution( - "block_db_txn_count", labels=["block_name"] +block_db_txn_count = metrics.register_counter( + "block_db_txn_count", labels=["block_name"], + alternative_names=( + metrics.name_prefix + "_block_db_txn_count:total", + ), ) -block_db_txn_duration = metrics.register_distribution( - "block_db_txn_duration", labels=["block_name"] +# seconds spent waiting for db txns, excluding scheduling time, in this block +block_db_txn_duration = metrics.register_counter( + "block_db_txn_duration_seconds", labels=["block_name"], + alternative_names=( + metrics.name_prefix + "_block_db_txn_duration:total", + ), +) + +# seconds spent waiting for a db connection, in this block +block_db_sched_duration = metrics.register_counter( + "block_db_sched_duration_seconds", labels=["block_name"], ) @@ -64,7 +101,9 @@ def measure_func(name): class Measure(object): __slots__ = [ "clock", "name", "start_context", "start", "new_context", "ru_utime", - "ru_stime", "db_txn_count", "db_txn_duration", "created_context" + "ru_stime", + "db_txn_count", "db_txn_duration_ms", "db_sched_duration_ms", + "created_context", ] def __init__(self, clock, name): @@ -84,13 +123,16 @@ class Measure(object): self.ru_utime, self.ru_stime = self.start_context.get_resource_usage() self.db_txn_count = self.start_context.db_txn_count - self.db_txn_duration = self.start_context.db_txn_duration + self.db_txn_duration_ms = self.start_context.db_txn_duration_ms + self.db_sched_duration_ms = self.start_context.db_sched_duration_ms def __exit__(self, exc_type, exc_val, exc_tb): if isinstance(exc_type, Exception) or not self.start_context: return duration = self.clock.time_msec() - self.start + + block_counter.inc(self.name) block_timer.inc_by(duration, self.name) context = LoggingContext.current_context() @@ -114,7 +156,12 @@ class Measure(object): context.db_txn_count - self.db_txn_count, self.name ) block_db_txn_duration.inc_by( - context.db_txn_duration - self.db_txn_duration, self.name + (context.db_txn_duration_ms - self.db_txn_duration_ms) / 1000., + self.name + ) + block_db_sched_duration.inc_by( + (context.db_sched_duration_ms - self.db_sched_duration_ms) / 1000., + self.name ) if self.created_context: diff --git a/synapse/util/retryutils.py b/synapse/util/retryutils.py index 1adedbb361..47b0bb5eb3 100644 --- a/synapse/util/retryutils.py +++ b/synapse/util/retryutils.py @@ -26,6 +26,18 @@ logger = logging.getLogger(__name__) class NotRetryingDestination(Exception): def __init__(self, retry_last_ts, retry_interval, destination): + """Raised by the limiter (and federation client) to indicate that we are + are deliberately not attempting to contact a given server. + + Args: + retry_last_ts (int): the unix ts in milliseconds of our last attempt + to contact the server. 0 indicates that the last attempt was + successful or that we've never actually attempted to connect. + retry_interval (int): the time in milliseconds to wait until the next + attempt. + destination (str): the domain in question + """ + msg = "Not retrying server %s." % (destination,) super(NotRetryingDestination, self).__init__(msg) diff --git a/synapse/util/threepids.py b/synapse/util/threepids.py new file mode 100644 index 0000000000..75efa0117b --- /dev/null +++ b/synapse/util/threepids.py @@ -0,0 +1,48 @@ +# -*- coding: utf-8 -*- +# Copyright 2018 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. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +import logging +import re + +logger = logging.getLogger(__name__) + + +def check_3pid_allowed(hs, medium, address): + """Checks whether a given format of 3PID is allowed to be used on this HS + + Args: + hs (synapse.server.HomeServer): server + medium (str): 3pid medium - e.g. email, msisdn + address (str): address within that medium (e.g. "wotan@matrix.org") + msisdns need to first have been canonicalised + Returns: + bool: whether the 3PID medium/address is allowed to be added to this HS + """ + + if hs.config.allowed_local_3pids: + for constraint in hs.config.allowed_local_3pids: + logger.debug( + "Checking 3PID %s (%s) against %s (%s)", + address, medium, constraint['pattern'], constraint['medium'], + ) + if ( + medium == constraint['medium'] and + re.match(constraint['pattern'], address) + ): + return True + else: + return True + + return False |