From 4cff617df1ba6f241fee6957cc44859f57edcc0e Mon Sep 17 00:00:00 2001 From: Erik Johnston Date: Wed, 25 Mar 2020 14:54:01 +0000 Subject: Move catchup of replication streams to worker. (#7024) This changes the replication protocol so that the server does not send down `RDATA` for rows that happened before the client connected. Instead, the server will send a `POSITION` and clients then query the database (or master out of band) to get up to date. --- tests/replication/tcp/streams/_base.py | 55 +++++++++++++++++++++++++--------- 1 file changed, 41 insertions(+), 14 deletions(-) (limited to 'tests/replication/tcp/streams/_base.py') diff --git a/tests/replication/tcp/streams/_base.py b/tests/replication/tcp/streams/_base.py index e96ad4ca4e..a755fe2879 100644 --- a/tests/replication/tcp/streams/_base.py +++ b/tests/replication/tcp/streams/_base.py @@ -12,6 +12,7 @@ # 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 mock import Mock from synapse.replication.tcp.commands import ReplicateCommand @@ -29,19 +30,37 @@ class BaseStreamTestCase(unittest.HomeserverTestCase): # build a replication server server_factory = ReplicationStreamProtocolFactory(self.hs) self.streamer = server_factory.streamer - server = server_factory.buildProtocol(None) + self.server = server_factory.buildProtocol(None) - # build a replication client, with a dummy handler - handler_factory = Mock() - self.test_handler = TestReplicationClientHandler() - self.test_handler.factory = handler_factory + self.test_handler = Mock(wraps=TestReplicationClientHandler()) self.client = ClientReplicationStreamProtocol( - "client", "test", clock, self.test_handler + hs, "client", "test", clock, self.test_handler, ) - # wire them together - self.client.makeConnection(FakeTransport(server, reactor)) - server.makeConnection(FakeTransport(self.client, reactor)) + self._client_transport = None + self._server_transport = None + + def reconnect(self): + if self._client_transport: + self.client.close() + + if self._server_transport: + self.server.close() + + self._client_transport = FakeTransport(self.server, self.reactor) + self.client.makeConnection(self._client_transport) + + self._server_transport = FakeTransport(self.client, self.reactor) + self.server.makeConnection(self._server_transport) + + def disconnect(self): + if self._client_transport: + self._client_transport = None + self.client.close() + + if self._server_transport: + self._server_transport = None + self.server.close() def replicate(self): """Tell the master side of replication that something has happened, and then @@ -50,19 +69,24 @@ class BaseStreamTestCase(unittest.HomeserverTestCase): self.streamer.on_notifier_poke() self.pump(0.1) - def replicate_stream(self, stream, token="NOW"): + def replicate_stream(self): """Make the client end a REPLICATE command to set up a subscription to a stream""" - self.client.send_command(ReplicateCommand(stream, token)) + self.client.send_command(ReplicateCommand()) class TestReplicationClientHandler(object): """Drop-in for ReplicationClientHandler which just collects RDATA rows""" def __init__(self): - self.received_rdata_rows = [] + self.streams = set() + self._received_rdata_rows = [] def get_streams_to_replicate(self): - return {} + positions = {s: 0 for s in self.streams} + for stream, token, _ in self._received_rdata_rows: + if stream in self.streams: + positions[stream] = max(token, positions.get(stream, 0)) + return positions def get_currently_syncing_users(self): return [] @@ -73,6 +97,9 @@ class TestReplicationClientHandler(object): def finished_connecting(self): pass + async def on_position(self, stream_name, token): + """Called when we get new position data.""" + async def on_rdata(self, stream_name, token, rows): for r in rows: - self.received_rdata_rows.append((stream_name, token, r)) + self._received_rdata_rows.append((stream_name, token, r)) -- cgit 1.5.1 From 5016b162fcf0372fe35404c64f80aeaf21461f31 Mon Sep 17 00:00:00 2001 From: Erik Johnston Date: Mon, 6 Apr 2020 09:58:42 +0100 Subject: Move client command handling out of TCP protocol (#7185) The aim here is to move the command handling out of the TCP protocol classes and to also merge the client and server command handling (so that we can reuse them for redis protocol). This PR simply moves the client paths to the new `ReplicationCommandHandler`, a future PR will move the server paths too. --- changelog.d/7185.misc | 1 + synapse/app/admin_cmd.py | 12 -- synapse/app/generic_worker.py | 9 +- synapse/replication/tcp/__init__.py | 30 ++- synapse/replication/tcp/client.py | 179 +++--------------- synapse/replication/tcp/handler.py | 252 +++++++++++++++++++++++++ synapse/replication/tcp/protocol.py | 197 +++---------------- synapse/server.py | 8 +- synapse/server.pyi | 7 +- tests/replication/slave/storage/_base.py | 15 +- tests/replication/tcp/streams/_base.py | 38 ++-- tests/replication/tcp/streams/test_receipts.py | 1 - 12 files changed, 378 insertions(+), 371 deletions(-) create mode 100644 changelog.d/7185.misc create mode 100644 synapse/replication/tcp/handler.py (limited to 'tests/replication/tcp/streams/_base.py') diff --git a/changelog.d/7185.misc b/changelog.d/7185.misc new file mode 100644 index 0000000000..deb9ca7021 --- /dev/null +++ b/changelog.d/7185.misc @@ -0,0 +1 @@ +Move client command handling out of TCP protocol. diff --git a/synapse/app/admin_cmd.py b/synapse/app/admin_cmd.py index 1c7c6ec0c8..a37818fe9a 100644 --- a/synapse/app/admin_cmd.py +++ b/synapse/app/admin_cmd.py @@ -43,7 +43,6 @@ from synapse.replication.slave.storage.push_rule import SlavedPushRuleStore 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.tcp.client import ReplicationClientHandler from synapse.server import HomeServer from synapse.util.logcontext import LoggingContext from synapse.util.versionstring import get_version_string @@ -79,17 +78,6 @@ class AdminCmdServer(HomeServer): def start_listening(self, listeners): pass - def build_tcp_replication(self): - return AdminCmdReplicationHandler(self) - - -class AdminCmdReplicationHandler(ReplicationClientHandler): - async def on_rdata(self, stream_name, token, rows): - pass - - def get_streams_to_replicate(self): - return {} - @defer.inlineCallbacks def export_data_command(hs, args): diff --git a/synapse/app/generic_worker.py b/synapse/app/generic_worker.py index 174bef360f..dcd0709a02 100644 --- a/synapse/app/generic_worker.py +++ b/synapse/app/generic_worker.py @@ -64,7 +64,7 @@ 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 SlavedTransactionStore -from synapse.replication.tcp.client import ReplicationClientHandler +from synapse.replication.tcp.client import ReplicationDataHandler from synapse.replication.tcp.commands import ClearUserSyncsCommand from synapse.replication.tcp.streams import ( AccountDataStream, @@ -603,7 +603,7 @@ class GenericWorkerServer(HomeServer): def remove_pusher(self, app_id, push_key, user_id): self.get_tcp_replication().send_remove_pusher(app_id, push_key, user_id) - def build_tcp_replication(self): + def build_replication_data_handler(self): return GenericWorkerReplicationHandler(self) def build_presence_handler(self): @@ -613,7 +613,7 @@ class GenericWorkerServer(HomeServer): return GenericWorkerTyping(self) -class GenericWorkerReplicationHandler(ReplicationClientHandler): +class GenericWorkerReplicationHandler(ReplicationDataHandler): def __init__(self, hs): super(GenericWorkerReplicationHandler, self).__init__(hs.get_datastore()) @@ -644,9 +644,6 @@ class GenericWorkerReplicationHandler(ReplicationClientHandler): args.update(self.send_handler.stream_positions()) return args - def get_currently_syncing_users(self): - return self.presence_handler.get_currently_syncing_users() - async def process_and_notify(self, stream_name, token, rows): try: if self.send_handler: diff --git a/synapse/replication/tcp/__init__.py b/synapse/replication/tcp/__init__.py index 81c2ea7ee9..523a1358d4 100644 --- a/synapse/replication/tcp/__init__.py +++ b/synapse/replication/tcp/__init__.py @@ -20,11 +20,31 @@ Further details can be found in docs/tcp_replication.rst Structure of the module: - * client.py - the client classes used for workers to connect to master + * handler.py - the classes used to handle sending/receiving commands to + replication * command.py - the definitions of all the valid commands - * protocol.py - contains bot the client and server protocol implementations, - these should not be used directly - * resource.py - the server classes that accepts and handle client connections - * streams.py - the definitons of all the valid streams + * protocol.py - the TCP protocol classes + * resource.py - handles streaming stream updates to replications + * streams/ - the definitons of all the valid streams + +The general interaction of the classes are: + + +---------------------+ + | ReplicationStreamer | + +---------------------+ + | + v + +---------------------------+ +----------------------+ + | ReplicationCommandHandler |---->|ReplicationDataHandler| + +---------------------------+ +----------------------+ + | ^ + v | + +-------------+ + | Protocols | + | (TCP/redis) | + +-------------+ + +Where the ReplicationDataHandler (or subclasses) handles incoming stream +updates. """ diff --git a/synapse/replication/tcp/client.py b/synapse/replication/tcp/client.py index e86d9805f1..700ae79158 100644 --- a/synapse/replication/tcp/client.py +++ b/synapse/replication/tcp/client.py @@ -16,26 +16,16 @@ """ import logging -from typing import Dict, List, Optional +from typing import TYPE_CHECKING, Dict -from twisted.internet import defer from twisted.internet.protocol import ReconnectingClientFactory from synapse.replication.slave.storage._base import BaseSlavedStore -from synapse.replication.tcp.protocol import ( - AbstractReplicationClientHandler, - ClientReplicationStreamProtocol, -) - -from .commands import ( - Command, - FederationAckCommand, - InvalidateCacheCommand, - RemoteServerUpCommand, - RemovePusherCommand, - UserIpCommand, - UserSyncCommand, -) +from synapse.replication.tcp.protocol import ClientReplicationStreamProtocol + +if TYPE_CHECKING: + from synapse.server import HomeServer + from synapse.replication.tcp.handler import ReplicationCommandHandler logger = logging.getLogger(__name__) @@ -44,16 +34,20 @@ class ReplicationClientFactory(ReconnectingClientFactory): """Factory for building connections to the master. Will reconnect if the connection is lost. - Accepts a handler that will be called when new data is available or data - is required. + Accepts a handler that is passed to `ClientReplicationStreamProtocol`. """ initialDelay = 0.1 maxDelay = 1 # Try at least once every N seconds - def __init__(self, hs, client_name, handler: AbstractReplicationClientHandler): + def __init__( + self, + hs: "HomeServer", + client_name: str, + command_handler: "ReplicationCommandHandler", + ): self.client_name = client_name - self.handler = handler + self.command_handler = command_handler self.server_name = hs.config.server_name self.hs = hs self._clock = hs.get_clock() # As self.clock is defined in super class @@ -66,7 +60,11 @@ class ReplicationClientFactory(ReconnectingClientFactory): def buildProtocol(self, addr): logger.info("Connected to replication: %r", addr) return ClientReplicationStreamProtocol( - self.hs, self.client_name, self.server_name, self._clock, self.handler, + self.hs, + self.client_name, + self.server_name, + self._clock, + self.command_handler, ) def clientConnectionLost(self, connector, reason): @@ -78,41 +76,17 @@ class ReplicationClientFactory(ReconnectingClientFactory): ReconnectingClientFactory.clientConnectionFailed(self, connector, reason) -class ReplicationClientHandler(AbstractReplicationClientHandler): - """A base handler that can be passed to the ReplicationClientFactory. +class ReplicationDataHandler: + """Handles incoming stream updates from replication. - By default proxies incoming replication data to the SlaveStore. + This instance notifies the slave data store about updates. Can be subclassed + to handle updates in additional ways. """ def __init__(self, store: BaseSlavedStore): self.store = store - # The current connection. None if we are currently (re)connecting - self.connection = None - - # Any pending commands to be sent once a new connection has been - # established - self.pending_commands = [] # type: List[Command] - - # Map from string -> deferred, to wake up when receiveing a SYNC with - # the given string. - # Used for tests. - self.awaiting_syncs = {} # type: Dict[str, defer.Deferred] - - # The factory used to create connections. - self.factory = None # type: Optional[ReplicationClientFactory] - - def start_replication(self, hs): - """Helper method to start a replication connection to the remote server - using TCP. - """ - client_name = hs.config.worker_name - self.factory = ReplicationClientFactory(hs, client_name, self) - host = hs.config.worker_replication_host - port = hs.config.worker_replication_port - hs.get_reactor().connectTCP(host, port, self.factory) - - async def on_rdata(self, stream_name, token, rows): + async def on_rdata(self, stream_name: str, token: int, rows: list): """Called to handle a batch of replication data with a given stream token. By default this just pokes the slave store. Can be overridden in subclasses to @@ -124,30 +98,8 @@ class ReplicationClientHandler(AbstractReplicationClientHandler): rows (list): a list of Stream.ROW_TYPE objects as returned by Stream.parse_row. """ - logger.debug("Received rdata %s -> %s", stream_name, token) self.store.process_replication_rows(stream_name, token, rows) - async def on_position(self, stream_name, token): - """Called when we get new position data. By default this just pokes - the slave store. - - Can be overriden in subclasses to handle more. - """ - self.store.process_replication_rows(stream_name, token, []) - - def on_sync(self, data): - """When we received a SYNC we wake up any deferreds that were waiting - for the sync with the given data. - - Used by tests. - """ - d = self.awaiting_syncs.pop(data, None) - if d: - d.callback(data) - - def on_remote_server_up(self, server: str): - """Called when get a new REMOTE_SERVER_UP command.""" - def get_streams_to_replicate(self) -> Dict[str, int]: """Called when a new connection has been established and we need to subscribe to streams. @@ -163,85 +115,10 @@ class ReplicationClientHandler(AbstractReplicationClientHandler): args["account_data"] = user_account_data elif room_account_data: args["account_data"] = room_account_data - return args - def get_currently_syncing_users(self): - """Get the list of currently syncing users (if any). This is called - when a connection has been established and we need to send the - currently syncing users. (Overriden by the synchrotron's only) - """ - return [] - - def send_command(self, cmd): - """Send a command to master (when we get establish a connection if we - don't have one already.) - """ - if self.connection: - self.connection.send_command(cmd) - else: - logger.warning("Queuing command as not connected: %r", cmd.NAME) - self.pending_commands.append(cmd) - - def send_federation_ack(self, token): - """Ack data for the federation stream. This allows the master to drop - data stored purely in memory. - """ - self.send_command(FederationAckCommand(token)) - - def send_user_sync(self, instance_id, user_id, is_syncing, last_sync_ms): - """Poke the master that a user has started/stopped syncing. - """ - self.send_command( - UserSyncCommand(instance_id, user_id, is_syncing, last_sync_ms) - ) - - def send_remove_pusher(self, app_id, push_key, user_id): - """Poke the master to remove a pusher for a user - """ - cmd = RemovePusherCommand(app_id, push_key, user_id) - self.send_command(cmd) - - def send_invalidate_cache(self, cache_func, keys): - """Poke the master to invalidate a cache. - """ - cmd = InvalidateCacheCommand(cache_func.__name__, keys) - self.send_command(cmd) - - def send_user_ip(self, user_id, access_token, ip, user_agent, device_id, last_seen): - """Tell the master that the user made a request. - """ - cmd = UserIpCommand(user_id, access_token, ip, user_agent, device_id, last_seen) - self.send_command(cmd) - - def send_remote_server_up(self, server: str): - self.send_command(RemoteServerUpCommand(server)) - - def await_sync(self, data): - """Returns a deferred that is resolved when we receive a SYNC command - with given data. - - [Not currently] used by tests. - """ - return self.awaiting_syncs.setdefault(data, defer.Deferred()) - - def update_connection(self, connection): - """Called when a connection has been established (or lost with None). - """ - self.connection = connection - if connection: - for cmd in self.pending_commands: - connection.send_command(cmd) - self.pending_commands = [] - - def finished_connecting(self): - """Called when we have successfully subscribed and caught up to all - streams we're interested in. - """ - logger.info("Finished connecting to server") + async def on_position(self, stream_name: str, token: int): + self.store.process_replication_rows(stream_name, token, []) - # We don't reset the delay any earlier as otherwise if there is a - # problem during start up we'll end up tight looping connecting to the - # server. - if self.factory: - self.factory.resetDelay() + def on_remote_server_up(self, server: str): + """Called when get a new REMOTE_SERVER_UP command.""" diff --git a/synapse/replication/tcp/handler.py b/synapse/replication/tcp/handler.py new file mode 100644 index 0000000000..12a1cfd6d1 --- /dev/null +++ b/synapse/replication/tcp/handler.py @@ -0,0 +1,252 @@ +# -*- coding: utf-8 -*- +# Copyright 2017 Vector Creations Ltd +# Copyright 2020 The Matrix.org Foundation C.I.C. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +import logging +from typing import Any, Callable, Dict, List, Optional, Set + +from prometheus_client import Counter + +from synapse.replication.tcp.client import ReplicationClientFactory +from synapse.replication.tcp.commands import ( + Command, + FederationAckCommand, + InvalidateCacheCommand, + PositionCommand, + RdataCommand, + RemoteServerUpCommand, + RemovePusherCommand, + SyncCommand, + UserIpCommand, + UserSyncCommand, +) +from synapse.replication.tcp.streams import STREAMS_MAP, Stream +from synapse.util.async_helpers import Linearizer + +logger = logging.getLogger(__name__) + + +# number of updates received for each RDATA stream +inbound_rdata_count = Counter( + "synapse_replication_tcp_protocol_inbound_rdata_count", "", ["stream_name"] +) + + +class ReplicationCommandHandler: + """Handles incoming commands from replication as well as sending commands + back out to connections. + """ + + def __init__(self, hs): + self._replication_data_handler = hs.get_replication_data_handler() + self._presence_handler = hs.get_presence_handler() + + # Set of streams that we've caught up with. + self._streams_connected = set() # type: Set[str] + + self._streams = { + stream.NAME: stream(hs) for stream in STREAMS_MAP.values() + } # type: Dict[str, Stream] + + self._position_linearizer = Linearizer("replication_position") + + # Map of stream to batched updates. See RdataCommand for info on how + # batching works. + self._pending_batches = {} # type: Dict[str, List[Any]] + + # The factory used to create connections. + self._factory = None # type: Optional[ReplicationClientFactory] + + # The current connection. None if we are currently (re)connecting + self._connection = None + + def start_replication(self, hs): + """Helper method to start a replication connection to the remote server + using TCP. + """ + client_name = hs.config.worker_name + self._factory = ReplicationClientFactory(hs, client_name, self) + host = hs.config.worker_replication_host + port = hs.config.worker_replication_port + hs.get_reactor().connectTCP(host, port, self._factory) + + async def on_RDATA(self, cmd: RdataCommand): + stream_name = cmd.stream_name + inbound_rdata_count.labels(stream_name).inc() + + try: + row = STREAMS_MAP[stream_name].parse_row(cmd.row) + except Exception: + logger.exception("Failed to parse RDATA: %r %r", stream_name, cmd.row) + raise + + if cmd.token is None or stream_name not in self._streams_connected: + # I.e. either this is part of a batch of updates for this stream (in + # which case batch until we get an update for the stream with a non + # None token) or we're currently connecting so we queue up rows. + 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(stream_name, []) + rows.append(row) + await self.on_rdata(stream_name, cmd.token, rows) + + async def on_rdata(self, stream_name: str, token: int, rows: list): + """Called to handle a batch of replication data with a given stream token. + + Args: + stream_name: name of the replication stream for this batch of rows + token: stream token for this batch of rows + rows: a list of Stream.ROW_TYPE objects as returned by + Stream.parse_row. + """ + logger.debug("Received rdata %s -> %s", stream_name, token) + await self._replication_data_handler.on_rdata(stream_name, token, rows) + + async def on_POSITION(self, cmd: PositionCommand): + stream = self._streams.get(cmd.stream_name) + if not stream: + logger.error("Got POSITION for unknown stream: %s", cmd.stream_name) + return + + # We protect catching up with a linearizer in case the replication + # connection reconnects under us. + with await self._position_linearizer.queue(cmd.stream_name): + # We're about to go and catch up with the stream, so mark as connecting + # to stop RDATA being handled at the same time by removing stream from + # list of connected streams. We also clear any batched up RDATA from + # before we got the POSITION. + self._streams_connected.discard(cmd.stream_name) + self._pending_batches.clear() + + # Find where we previously streamed up to. + current_token = self._replication_data_handler.get_streams_to_replicate().get( + cmd.stream_name + ) + if current_token is None: + logger.warning( + "Got POSITION for stream we're not subscribed to: %s", + cmd.stream_name, + ) + return + + # Fetch all updates between then and now. + limited = True + while limited: + updates, current_token, limited = await stream.get_updates_since( + current_token, cmd.token + ) + if updates: + await self.on_rdata( + cmd.stream_name, + current_token, + [stream.parse_row(update[1]) for update in updates], + ) + + # We've now caught up to position sent to us, notify handler. + await self._replication_data_handler.on_position(cmd.stream_name, cmd.token) + + # Handle any RDATA that came in while we were catching up. + rows = self._pending_batches.pop(cmd.stream_name, []) + if rows: + await self._replication_data_handler.on_rdata( + cmd.stream_name, rows[-1].token, rows + ) + + self._streams_connected.add(cmd.stream_name) + + async def on_SYNC(self, cmd: SyncCommand): + pass + + async def on_REMOTE_SERVER_UP(self, cmd: RemoteServerUpCommand): + """"Called when get a new REMOTE_SERVER_UP command.""" + self._replication_data_handler.on_remote_server_up(cmd.data) + + def get_currently_syncing_users(self): + """Get the list of currently syncing users (if any). This is called + when a connection has been established and we need to send the + currently syncing users. + """ + return self._presence_handler.get_currently_syncing_users() + + def update_connection(self, connection): + """Called when a connection has been established (or lost with None). + """ + self._connection = connection + + def finished_connecting(self): + """Called when we have successfully subscribed and caught up to all + streams we're interested in. + """ + logger.info("Finished connecting to server") + + # We don't reset the delay any earlier as otherwise if there is a + # problem during start up we'll end up tight looping connecting to the + # server. + if self._factory: + self._factory.resetDelay() + + def send_command(self, cmd: Command): + """Send a command to master (when we get establish a connection if we + don't have one already.) + """ + if self._connection: + self._connection.send_command(cmd) + else: + logger.warning("Dropping command as not connected: %r", cmd.NAME) + + def send_federation_ack(self, token: int): + """Ack data for the federation stream. This allows the master to drop + data stored purely in memory. + """ + self.send_command(FederationAckCommand(token)) + + def send_user_sync( + self, instance_id: str, user_id: str, is_syncing: bool, last_sync_ms: int + ): + """Poke the master that a user has started/stopped syncing. + """ + self.send_command( + UserSyncCommand(instance_id, user_id, is_syncing, last_sync_ms) + ) + + def send_remove_pusher(self, app_id: str, push_key: str, user_id: str): + """Poke the master to remove a pusher for a user + """ + cmd = RemovePusherCommand(app_id, push_key, user_id) + self.send_command(cmd) + + def send_invalidate_cache(self, cache_func: Callable, keys: tuple): + """Poke the master to invalidate a cache. + """ + cmd = InvalidateCacheCommand(cache_func.__name__, keys) + self.send_command(cmd) + + def send_user_ip( + self, + user_id: str, + access_token: str, + ip: str, + user_agent: str, + device_id: str, + last_seen: int, + ): + """Tell the master that the user made a request. + """ + cmd = UserIpCommand(user_id, access_token, ip, user_agent, device_id, last_seen) + self.send_command(cmd) + + def send_remote_server_up(self, server: str): + self.send_command(RemoteServerUpCommand(server)) diff --git a/synapse/replication/tcp/protocol.py b/synapse/replication/tcp/protocol.py index dae246825f..f2a37f568e 100644 --- a/synapse/replication/tcp/protocol.py +++ b/synapse/replication/tcp/protocol.py @@ -46,12 +46,11 @@ indicate which side is sending, these are *not* included on the wire:: > ERROR server stopping * connection closed by server * """ -import abc import fcntl import logging import struct from collections import defaultdict -from typing import Any, DefaultDict, Dict, List, Set +from typing import TYPE_CHECKING, DefaultDict, List from six import iteritems @@ -78,13 +77,12 @@ from synapse.replication.tcp.commands import ( SyncCommand, UserSyncCommand, ) -from synapse.replication.tcp.streams import STREAMS_MAP, Stream from synapse.types import Collection from synapse.util import Clock from synapse.util.stringutils import random_string -MYPY = False -if MYPY: +if TYPE_CHECKING: + from synapse.replication.tcp.handler import ReplicationCommandHandler from synapse.server import HomeServer @@ -475,71 +473,6 @@ class ServerReplicationStreamProtocol(BaseReplicationStreamProtocol): self.streamer.lost_connection(self) -class AbstractReplicationClientHandler(metaclass=abc.ABCMeta): - """ - The interface for the handler that should be passed to - ClientReplicationStreamProtocol - """ - - @abc.abstractmethod - async def on_rdata(self, stream_name, token, rows): - """Called to handle a batch of replication data with a given stream token. - - Args: - stream_name (str): name of the replication stream for this batch of rows - token (int): stream token for this batch of rows - rows (list): a list of Stream.ROW_TYPE objects as returned by - Stream.parse_row. - """ - raise NotImplementedError() - - @abc.abstractmethod - async def on_position(self, stream_name, token): - """Called when we get new position data.""" - raise NotImplementedError() - - @abc.abstractmethod - def on_sync(self, data): - """Called when get a new SYNC command.""" - raise NotImplementedError() - - @abc.abstractmethod - async def on_remote_server_up(self, server: str): - """Called when get a new REMOTE_SERVER_UP command.""" - raise NotImplementedError() - - @abc.abstractmethod - def get_streams_to_replicate(self): - """Called when a new connection has been established and we need to - subscribe to streams. - - Returns: - map from stream name to the most recent update we have for - that stream (ie, the point we want to start replicating from) - """ - raise NotImplementedError() - - @abc.abstractmethod - def get_currently_syncing_users(self): - """Get the list of currently syncing users (if any). This is called - when a connection has been established and we need to send the - currently syncing users.""" - raise NotImplementedError() - - @abc.abstractmethod - def update_connection(self, connection): - """Called when a connection has been established (or lost with None). - """ - raise NotImplementedError() - - @abc.abstractmethod - def finished_connecting(self): - """Called when we have successfully subscribed and caught up to all - streams we're interested in. - """ - raise NotImplementedError() - - class ClientReplicationStreamProtocol(BaseReplicationStreamProtocol): VALID_INBOUND_COMMANDS = VALID_SERVER_COMMANDS VALID_OUTBOUND_COMMANDS = VALID_CLIENT_COMMANDS @@ -550,7 +483,7 @@ class ClientReplicationStreamProtocol(BaseReplicationStreamProtocol): client_name: str, server_name: str, clock: Clock, - handler: AbstractReplicationClientHandler, + command_handler: "ReplicationCommandHandler", ): BaseReplicationStreamProtocol.__init__(self, clock) @@ -558,20 +491,7 @@ class ClientReplicationStreamProtocol(BaseReplicationStreamProtocol): self.client_name = client_name self.server_name = server_name - self.handler = handler - - self.streams = { - stream.NAME: stream(hs) for stream in STREAMS_MAP.values() - } # type: Dict[str, Stream] - - # Set of stream names that have been subscribe to, but haven't yet - # caught up with. This is used to track when the client has been fully - # connected to the remote. - self.streams_connecting = set(STREAMS_MAP) # type: Set[str] - - # Map of stream to batched updates. See RdataCommand for info on how - # batching works. - self.pending_batches = {} # type: Dict[str, List[Any]] + self.handler = command_handler def connectionMade(self): self.send_command(NameCommand(self.client_name)) @@ -589,89 +509,39 @@ class ClientReplicationStreamProtocol(BaseReplicationStreamProtocol): # We've now finished connecting to so inform the client handler self.handler.update_connection(self) + self.handler.finished_connecting() - async def on_SERVER(self, cmd): - if cmd.data != self.server_name: - logger.error("[%s] Connected to wrong remote: %r", self.id(), cmd.data) - self.send_error("Wrong remote") - - async def on_RDATA(self, cmd): - stream_name = cmd.stream_name - inbound_rdata_count.labels(stream_name).inc() - - try: - row = STREAMS_MAP[stream_name].parse_row(cmd.row) - except Exception: - logger.exception( - "[%s] Failed to parse RDATA: %r %r", self.id(), stream_name, cmd.row - ) - raise - - if cmd.token is None or stream_name in self.streams_connecting: - # 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(stream_name, []).append(row) - else: - # Check if this is the last of a batch of updates - rows = self.pending_batches.pop(stream_name, []) - rows.append(row) - await self.handler.on_rdata(stream_name, cmd.token, rows) - - async def on_POSITION(self, cmd: PositionCommand): - stream = self.streams.get(cmd.stream_name) - if not stream: - logger.error("Got POSITION for unknown stream: %s", cmd.stream_name) - return - - # Find where we previously streamed up to. - current_token = self.handler.get_streams_to_replicate().get(cmd.stream_name) - if current_token is None: - logger.warning( - "Got POSITION for stream we're not subscribed to: %s", cmd.stream_name - ) - return - - # Fetch all updates between then and now. - limited = True - while limited: - updates, current_token, limited = await stream.get_updates_since( - current_token, cmd.token - ) - - # Check if the connection was closed underneath us, if so we bail - # rather than risk having concurrent catch ups going on. - if self.state == ConnectionStates.CLOSED: - return - - if updates: - await self.handler.on_rdata( - cmd.stream_name, - current_token, - [stream.parse_row(update[1]) for update in updates], - ) + async def handle_command(self, cmd: Command): + """Handle a command we have received over the replication stream. - # We've now caught up to position sent to us, notify handler. - await self.handler.on_position(cmd.stream_name, cmd.token) + Delegates to `command_handler.on_`, which must return an + awaitable. - self.streams_connecting.discard(cmd.stream_name) - if not self.streams_connecting: - self.handler.finished_connecting() + Args: + cmd: received command + """ + handled = False - # Check if the connection was closed underneath us, if so we bail - # rather than risk having concurrent catch ups going on. - if self.state == ConnectionStates.CLOSED: - return + # First call any command handlers on this instance. These are for TCP + # specific handling. + cmd_func = getattr(self, "on_%s" % (cmd.NAME,), None) + if cmd_func: + await cmd_func(cmd) + handled = True - # Handle any RDATA that came in while we were catching up. - rows = self.pending_batches.pop(cmd.stream_name, []) - if rows: - await self.handler.on_rdata(cmd.stream_name, rows[-1].token, rows) + # Then call out to the handler. + cmd_func = getattr(self.handler, "on_%s" % (cmd.NAME,), None) + if cmd_func: + await cmd_func(cmd) + handled = True - async def on_SYNC(self, cmd): - self.handler.on_sync(cmd.data) + if not handled: + logger.warning("Unhandled command: %r", cmd) - async def on_REMOTE_SERVER_UP(self, cmd: RemoteServerUpCommand): - self.handler.on_remote_server_up(cmd.data) + async def on_SERVER(self, cmd): + if cmd.data != self.server_name: + logger.error("[%s] Connected to wrong remote: %r", self.id(), cmd.data) + self.send_error("Wrong remote") def replicate(self): """Send the subscription request to the server @@ -768,8 +638,3 @@ tcp_outbound_commands = LaterGauge( for k, count in iteritems(p.outbound_commands_counter) }, ) - -# number of updates received for each RDATA stream -inbound_rdata_count = Counter( - "synapse_replication_tcp_protocol_inbound_rdata_count", "", ["stream_name"] -) diff --git a/synapse/server.py b/synapse/server.py index 9228e1c892..9d273c980c 100644 --- a/synapse/server.py +++ b/synapse/server.py @@ -87,6 +87,8 @@ from synapse.http.matrixfederationclient import MatrixFederationHttpClient from synapse.notifier import Notifier from synapse.push.action_generator import ActionGenerator from synapse.push.pusherpool import PusherPool +from synapse.replication.tcp.client import ReplicationDataHandler +from synapse.replication.tcp.handler import ReplicationCommandHandler from synapse.replication.tcp.resource import ReplicationStreamer from synapse.rest.media.v1.media_repository import ( MediaRepository, @@ -206,6 +208,7 @@ class HomeServer(object): "password_policy_handler", "storage", "replication_streamer", + "replication_data_handler", ] REQUIRED_ON_MASTER_STARTUP = ["user_directory_handler", "stats_handler"] @@ -468,7 +471,7 @@ class HomeServer(object): return ReadMarkerHandler(self) def build_tcp_replication(self): - raise NotImplementedError() + return ReplicationCommandHandler(self) def build_action_generator(self): return ActionGenerator(self) @@ -562,6 +565,9 @@ class HomeServer(object): def build_replication_streamer(self) -> ReplicationStreamer: return ReplicationStreamer(self) + def build_replication_data_handler(self): + return ReplicationDataHandler(self.get_datastore()) + 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 9d1dfa71e7..9013e9bac9 100644 --- a/synapse/server.pyi +++ b/synapse/server.pyi @@ -19,6 +19,7 @@ import synapse.handlers.set_password import synapse.http.client import synapse.notifier import synapse.replication.tcp.client +import synapse.replication.tcp.handler import synapse.rest.media.v1.media_repository import synapse.server_notices.server_notices_manager import synapse.server_notices.server_notices_sender @@ -106,7 +107,11 @@ class HomeServer(object): pass def get_tcp_replication( self, - ) -> synapse.replication.tcp.client.ReplicationClientHandler: + ) -> synapse.replication.tcp.handler.ReplicationCommandHandler: + pass + def get_replication_data_handler( + self, + ) -> synapse.replication.tcp.client.ReplicationDataHandler: pass def get_federation_registry( self, diff --git a/tests/replication/slave/storage/_base.py b/tests/replication/slave/storage/_base.py index 2a1e7c7166..8902a5ab69 100644 --- a/tests/replication/slave/storage/_base.py +++ b/tests/replication/slave/storage/_base.py @@ -17,8 +17,9 @@ from mock import Mock, NonCallableMock from synapse.replication.tcp.client import ( ReplicationClientFactory, - ReplicationClientHandler, + ReplicationDataHandler, ) +from synapse.replication.tcp.handler import ReplicationCommandHandler from synapse.replication.tcp.resource import ReplicationStreamProtocolFactory from synapse.storage.database import make_conn @@ -51,15 +52,19 @@ class BaseSlavedStoreTestCase(unittest.HomeserverTestCase): self.event_id = 0 server_factory = ReplicationStreamProtocolFactory(self.hs) - self.streamer = server_factory.streamer + self.streamer = hs.get_replication_streamer() - handler_factory = Mock() - self.replication_handler = ReplicationClientHandler(self.slaved_store) - self.replication_handler.factory = handler_factory + # We now do some gut wrenching so that we have a client that is based + # off of the slave store rather than the main store. + self.replication_handler = ReplicationCommandHandler(self.hs) + self.replication_handler._replication_data_handler = ReplicationDataHandler( + self.slaved_store + ) client_factory = ReplicationClientFactory( self.hs, "client_name", self.replication_handler ) + client_factory.handler = self.replication_handler server = server_factory.buildProtocol(None) client = client_factory.buildProtocol(None) diff --git a/tests/replication/tcp/streams/_base.py b/tests/replication/tcp/streams/_base.py index a755fe2879..32238fe79a 100644 --- a/tests/replication/tcp/streams/_base.py +++ b/tests/replication/tcp/streams/_base.py @@ -15,7 +15,7 @@ from mock import Mock -from synapse.replication.tcp.commands import ReplicateCommand +from synapse.replication.tcp.handler import ReplicationCommandHandler from synapse.replication.tcp.protocol import ClientReplicationStreamProtocol from synapse.replication.tcp.resource import ReplicationStreamProtocolFactory @@ -26,15 +26,20 @@ from tests.server import FakeTransport class BaseStreamTestCase(unittest.HomeserverTestCase): """Base class for tests of the replication streams""" + def make_homeserver(self, reactor, clock): + self.test_handler = Mock(wraps=TestReplicationDataHandler()) + return self.setup_test_homeserver(replication_data_handler=self.test_handler) + def prepare(self, reactor, clock, hs): # build a replication server - server_factory = ReplicationStreamProtocolFactory(self.hs) - self.streamer = server_factory.streamer + server_factory = ReplicationStreamProtocolFactory(hs) + self.streamer = hs.get_replication_streamer() self.server = server_factory.buildProtocol(None) - self.test_handler = Mock(wraps=TestReplicationClientHandler()) + repl_handler = ReplicationCommandHandler(hs) + repl_handler.handler = self.test_handler self.client = ClientReplicationStreamProtocol( - hs, "client", "test", clock, self.test_handler, + hs, "client", "test", clock, repl_handler, ) self._client_transport = None @@ -69,13 +74,9 @@ class BaseStreamTestCase(unittest.HomeserverTestCase): self.streamer.on_notifier_poke() self.pump(0.1) - def replicate_stream(self): - """Make the client end a REPLICATE command to set up a subscription to a stream""" - self.client.send_command(ReplicateCommand()) - -class TestReplicationClientHandler(object): - """Drop-in for ReplicationClientHandler which just collects RDATA rows""" +class TestReplicationDataHandler: + """Drop-in for ReplicationDataHandler which just collects RDATA rows""" def __init__(self): self.streams = set() @@ -88,18 +89,9 @@ class TestReplicationClientHandler(object): positions[stream] = max(token, positions.get(stream, 0)) return positions - def get_currently_syncing_users(self): - return [] - - def update_connection(self, connection): - pass - - def finished_connecting(self): - pass - - async def on_position(self, stream_name, token): - """Called when we get new position data.""" - async def on_rdata(self, stream_name, token, rows): for r in rows: self._received_rdata_rows.append((stream_name, token, r)) + + async def on_position(self, stream_name, token): + pass diff --git a/tests/replication/tcp/streams/test_receipts.py b/tests/replication/tcp/streams/test_receipts.py index 0ec0825a0e..a0206f7363 100644 --- a/tests/replication/tcp/streams/test_receipts.py +++ b/tests/replication/tcp/streams/test_receipts.py @@ -24,7 +24,6 @@ class ReceiptsStreamTestCase(BaseStreamTestCase): self.reconnect() # make the client subscribe to the receipts stream - self.replicate_stream() self.test_handler.streams.add("receipts") # tell the master to send a new receipt -- cgit 1.5.1 From fce663889b150e55b19097dd9b7fed7aca8abccc Mon Sep 17 00:00:00 2001 From: Erik Johnston Date: Tue, 28 Apr 2020 17:42:03 +0100 Subject: Add some replication tests (#7278) Specifically some tests for the typing stream, which means we test streams that fetch missing updates via HTTP (rather than via the DB). We also shuffle things around a bit so that we create two separate `HomeServer` objects, rather than trying to insert a slaved store into places. Note: `test_typing.py` is heavily inspired by `test_receipts.py` --- changelog.d/7278.misc | 1 + tests/replication/tcp/streams/_base.py | 229 +++++++++++++++++++++++++-- tests/replication/tcp/streams/test_typing.py | 80 ++++++++++ 3 files changed, 299 insertions(+), 11 deletions(-) create mode 100644 changelog.d/7278.misc create mode 100644 tests/replication/tcp/streams/test_typing.py (limited to 'tests/replication/tcp/streams/_base.py') diff --git a/changelog.d/7278.misc b/changelog.d/7278.misc new file mode 100644 index 0000000000..8a4c4328f4 --- /dev/null +++ b/changelog.d/7278.misc @@ -0,0 +1 @@ +Add some unit tests for replication. diff --git a/tests/replication/tcp/streams/_base.py b/tests/replication/tcp/streams/_base.py index 32238fe79a..82f15c64e0 100644 --- a/tests/replication/tcp/streams/_base.py +++ b/tests/replication/tcp/streams/_base.py @@ -12,34 +12,67 @@ # 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 typing import Optional from mock import Mock +import attr + +from twisted.internet.interfaces import IConsumer, IPullProducer, IReactorTime +from twisted.internet.task import LoopingCall +from twisted.web.http import HTTPChannel + +from synapse.app.generic_worker import GenericWorkerServer +from synapse.http.site import SynapseRequest +from synapse.replication.tcp.client import ReplicationDataHandler from synapse.replication.tcp.handler import ReplicationCommandHandler from synapse.replication.tcp.protocol import ClientReplicationStreamProtocol from synapse.replication.tcp.resource import ReplicationStreamProtocolFactory +from synapse.util import Clock from tests import unittest from tests.server import FakeTransport +logger = logging.getLogger(__name__) + class BaseStreamTestCase(unittest.HomeserverTestCase): """Base class for tests of the replication streams""" - def make_homeserver(self, reactor, clock): - self.test_handler = Mock(wraps=TestReplicationDataHandler()) - return self.setup_test_homeserver(replication_data_handler=self.test_handler) - def prepare(self, reactor, clock, hs): # build a replication server server_factory = ReplicationStreamProtocolFactory(hs) self.streamer = hs.get_replication_streamer() self.server = server_factory.buildProtocol(None) - repl_handler = ReplicationCommandHandler(hs) - repl_handler.handler = self.test_handler + # Make a new HomeServer object for the worker + config = self.default_config() + config["worker_app"] = "synapse.app.generic_worker" + config["worker_replication_host"] = "testserv" + config["worker_replication_http_port"] = "8765" + + self.reactor.lookups["testserv"] = "1.2.3.4" + + self.worker_hs = self.setup_test_homeserver( + http_client=None, + homeserverToUse=GenericWorkerServer, + config=config, + reactor=self.reactor, + ) + + # Since we use sqlite in memory databases we need to make sure the + # databases objects are the same. + self.worker_hs.get_datastore().db = hs.get_datastore().db + + self.test_handler = Mock( + wraps=TestReplicationDataHandler(self.worker_hs.get_datastore()) + ) + self.worker_hs.replication_data_handler = self.test_handler + + repl_handler = ReplicationCommandHandler(self.worker_hs) self.client = ClientReplicationStreamProtocol( - hs, "client", "test", clock, repl_handler, + self.worker_hs, "client", "test", clock, repl_handler, ) self._client_transport = None @@ -74,11 +107,75 @@ class BaseStreamTestCase(unittest.HomeserverTestCase): self.streamer.on_notifier_poke() self.pump(0.1) + def handle_http_replication_attempt(self) -> SynapseRequest: + """Asserts that a connection attempt was made to the master HS on the + HTTP replication port, then proxies it to the master HS object to be + handled. + + Returns: + The request object received by master HS. + """ + + # We should have an outbound connection attempt. + clients = self.reactor.tcpClients + self.assertEqual(len(clients), 1) + (host, port, client_factory, _timeout, _bindAddress) = clients.pop(0) + self.assertEqual(host, "1.2.3.4") + self.assertEqual(port, 8765) + + # Set up client side protocol + client_protocol = client_factory.buildProtocol(None) + + request_factory = OneShotRequestFactory() -class TestReplicationDataHandler: + # Set up the server side protocol + channel = _PushHTTPChannel(self.reactor) + channel.requestFactory = request_factory + channel.site = self.site + + # Connect client to server and vice versa. + client_to_server_transport = FakeTransport( + channel, self.reactor, client_protocol + ) + client_protocol.makeConnection(client_to_server_transport) + + server_to_client_transport = FakeTransport( + client_protocol, self.reactor, channel + ) + channel.makeConnection(server_to_client_transport) + + # The request will now be processed by `self.site` and the response + # streamed back. + self.reactor.advance(0) + + # We tear down the connection so it doesn't get reused without our + # knowledge. + server_to_client_transport.loseConnection() + client_to_server_transport.loseConnection() + + return request_factory.request + + def assert_request_is_get_repl_stream_updates( + self, request: SynapseRequest, stream_name: str + ): + """Asserts that the given request is a HTTP replication request for + fetching updates for given stream. + """ + + self.assertRegex( + request.path, + br"^/_synapse/replication/get_repl_stream_updates/%s/[^/]+$" + % (stream_name.encode("ascii"),), + ) + + self.assertEqual(request.method, b"GET") + + +class TestReplicationDataHandler(ReplicationDataHandler): """Drop-in for ReplicationDataHandler which just collects RDATA rows""" - def __init__(self): + def __init__(self, hs): + super().__init__(hs) self.streams = set() self._received_rdata_rows = [] @@ -90,8 +187,118 @@ class TestReplicationDataHandler: return positions async def on_rdata(self, stream_name, token, rows): + await super().on_rdata(stream_name, token, rows) for r in rows: self._received_rdata_rows.append((stream_name, token, r)) - async def on_position(self, stream_name, token): - pass + +@attr.s() +class OneShotRequestFactory: + """A simple request factory that generates a single `SynapseRequest` and + stores it for future use. Can only be used once. + """ + + request = attr.ib(default=None) + + def __call__(self, *args, **kwargs): + assert self.request is None + + self.request = SynapseRequest(*args, **kwargs) + return self.request + + +class _PushHTTPChannel(HTTPChannel): + """A HTTPChannel that wraps pull producers to push producers. + + This is a hack to get around the fact that HTTPChannel transparently wraps a + pull producer (which is what Synapse uses to reply to requests) with + `_PullToPush` to convert it to a push producer. Unfortunately `_PullToPush` + uses the standard reactor rather than letting us use our test reactor, which + makes it very hard to test. + """ + + def __init__(self, reactor: IReactorTime): + super().__init__() + self.reactor = reactor + + self._pull_to_push_producer = None + + def registerProducer(self, producer, streaming): + # Convert pull producers to push producer. + if not streaming: + self._pull_to_push_producer = _PullToPushProducer( + self.reactor, producer, self + ) + producer = self._pull_to_push_producer + + super().registerProducer(producer, True) + + def unregisterProducer(self): + if self._pull_to_push_producer: + # We need to manually stop the _PullToPushProducer. + self._pull_to_push_producer.stop() + + +class _PullToPushProducer: + """A push producer that wraps a pull producer. + """ + + def __init__( + self, reactor: IReactorTime, producer: IPullProducer, consumer: IConsumer + ): + self._clock = Clock(reactor) + self._producer = producer + self._consumer = consumer + + # While running we use a looping call with a zero delay to call + # resumeProducing on given producer. + self._looping_call = None # type: Optional[LoopingCall] + + # We start writing next reactor tick. + self._start_loop() + + def _start_loop(self): + """Start the looping call to + """ + + if not self._looping_call: + # Start a looping call which runs every tick. + self._looping_call = self._clock.looping_call(self._run_once, 0) + + def stop(self): + """Stops calling resumeProducing. + """ + if self._looping_call: + self._looping_call.stop() + self._looping_call = None + + def pauseProducing(self): + """Implements IPushProducer + """ + self.stop() + + def resumeProducing(self): + """Implements IPushProducer + """ + self._start_loop() + + def stopProducing(self): + """Implements IPushProducer + """ + self.stop() + self._producer.stopProducing() + + def _run_once(self): + """Calls resumeProducing on producer once. + """ + + try: + self._producer.resumeProducing() + except Exception: + logger.exception("Failed to call resumeProducing") + try: + self._consumer.unregisterProducer() + except Exception: + pass + + self.stopProducing() diff --git a/tests/replication/tcp/streams/test_typing.py b/tests/replication/tcp/streams/test_typing.py new file mode 100644 index 0000000000..f0ad6402ae --- /dev/null +++ b/tests/replication/tcp/streams/test_typing.py @@ -0,0 +1,80 @@ +# -*- coding: utf-8 -*- +# Copyright 2020 The Matrix.org Foundation C.I.C. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. +from synapse.handlers.typing import RoomMember +from synapse.replication.http import streams +from synapse.replication.tcp.streams import TypingStream + +from tests.replication.tcp.streams._base import BaseStreamTestCase + +USER_ID = "@feeling:blue" + + +class TypingStreamTestCase(BaseStreamTestCase): + servlets = [ + streams.register_servlets, + ] + + def test_typing(self): + typing = self.hs.get_typing_handler() + + room_id = "!bar:blue" + + self.reconnect() + + # make the client subscribe to the receipts stream + self.test_handler.streams.add("typing") + + typing._push_update(member=RoomMember(room_id, USER_ID), typing=True) + + self.reactor.advance(0) + + # We should now see an attempt to connect to the master + request = self.handle_http_replication_attempt() + self.assert_request_is_get_repl_stream_updates(request, "typing") + + self.test_handler.on_rdata.assert_called_once() + stream_name, token, rdata_rows = self.test_handler.on_rdata.call_args[0] + self.assertEqual(stream_name, "typing") + self.assertEqual(1, len(rdata_rows)) + row = rdata_rows[0] # type: TypingStream.TypingStreamRow + self.assertEqual(room_id, row.room_id) + self.assertEqual([USER_ID], row.user_ids) + + # Now let's disconnect and insert some data. + self.disconnect() + + self.test_handler.on_rdata.reset_mock() + + typing._push_update(member=RoomMember(room_id, USER_ID), typing=False) + + self.test_handler.on_rdata.assert_not_called() + + self.reconnect() + self.pump(0.1) + + # We should now see an attempt to connect to the master + request = self.handle_http_replication_attempt() + self.assert_request_is_get_repl_stream_updates(request, "typing") + + # The from token should be the token from the last RDATA we got. + self.assertEqual(int(request.args[b"from_token"][0]), token) + + self.test_handler.on_rdata.assert_called_once() + stream_name, token, rdata_rows = self.test_handler.on_rdata.call_args[0] + self.assertEqual(stream_name, "typing") + self.assertEqual(1, len(rdata_rows)) + row = rdata_rows[0] # type: TypingStream.TypingStreamRow + self.assertEqual(room_id, row.room_id) + self.assertEqual([], row.user_ids) -- cgit 1.5.1 From c2e1a2110fbe9ead26b4ecbb1afd504ed035a04d Mon Sep 17 00:00:00 2001 From: Richard van der Hoff <1389908+richvdh@users.noreply.github.com> Date: Wed, 29 Apr 2020 12:30:36 +0100 Subject: Fix limit logic for EventsStream (#7358) * Factor out functions for injecting events into database I want to add some more flexibility to the tools for injecting events into the database, and I don't want to clutter up HomeserverTestCase with them, so let's factor them out to a new file. * Rework TestReplicationDataHandler This wasn't very easy to work with: the mock wrapping was largely superfluous, and it's useful to be able to inspect the received rows, and clear out the received list. * Fix AssertionErrors being thrown by EventsStream Part of the problem was that there was an off-by-one error in the assertion, but also the limit logic was too simple. Fix it all up and add some tests. --- changelog.d/7358.bugfix | 1 + synapse/replication/tcp/handler.py | 4 +- synapse/replication/tcp/streams/events.py | 22 +- synapse/server.pyi | 5 + synapse/storage/data_stores/main/events_worker.py | 64 +++- tests/replication/tcp/streams/_base.py | 41 ++- tests/replication/tcp/streams/test_events.py | 417 ++++++++++++++++++++++ tests/replication/tcp/streams/test_receipts.py | 10 +- tests/replication/tcp/streams/test_typing.py | 11 +- tests/rest/client/v1/utils.py | 2 +- tests/test_utils/__init__.py | 20 ++ tests/test_utils/event_injection.py | 96 +++++ tests/unittest.py | 30 +- tox.ini | 2 + 14 files changed, 658 insertions(+), 67 deletions(-) create mode 100644 changelog.d/7358.bugfix create mode 100644 tests/replication/tcp/streams/test_events.py create mode 100644 tests/test_utils/event_injection.py (limited to 'tests/replication/tcp/streams/_base.py') diff --git a/changelog.d/7358.bugfix b/changelog.d/7358.bugfix new file mode 100644 index 0000000000..f49c600173 --- /dev/null +++ b/changelog.d/7358.bugfix @@ -0,0 +1 @@ +Fix a bug where event updates might not be sent over replication to worker processes after the stream falls behind. diff --git a/synapse/replication/tcp/handler.py b/synapse/replication/tcp/handler.py index 0db5a3a24d..3a8c7c7e2d 100644 --- a/synapse/replication/tcp/handler.py +++ b/synapse/replication/tcp/handler.py @@ -87,7 +87,9 @@ class ReplicationCommandHandler: stream.NAME: stream(hs) for stream in STREAMS_MAP.values() } # type: Dict[str, Stream] - self._position_linearizer = Linearizer("replication_position") + self._position_linearizer = Linearizer( + "replication_position", clock=self._clock + ) # Map of stream to batched updates. See RdataCommand for info on how # batching works. diff --git a/synapse/replication/tcp/streams/events.py b/synapse/replication/tcp/streams/events.py index aa50492569..52df81b1bd 100644 --- a/synapse/replication/tcp/streams/events.py +++ b/synapse/replication/tcp/streams/events.py @@ -170,22 +170,16 @@ class EventsStream(Stream): limited = False upper_limit = current_token - # next up is the state delta table - - state_rows = await self._store.get_all_updated_current_state_deltas( + # next up is the state delta table. + ( + state_rows, + upper_limit, + state_rows_limited, + ) = await self._store.get_all_updated_current_state_deltas( from_token, upper_limit, target_row_count - ) # type: List[Tuple] - - # again, if we've hit the limit there, we'll need to limit the other sources - assert len(state_rows) < target_row_count - if len(state_rows) == target_row_count: - assert state_rows[-1][0] <= upper_limit - upper_limit = state_rows[-1][0] - limited = True + ) - # FIXME: is it a given that there is only one row per stream_id in the - # state_deltas table (so that we can be sure that we have got all of the - # rows for upper_limit)? + limited = limited or state_rows_limited # finally, fetch the ex-outliers rows. We assume there are few enough of these # not to bother with the limit. diff --git a/synapse/server.pyi b/synapse/server.pyi index f1a5717028..fc5886f762 100644 --- a/synapse/server.pyi +++ b/synapse/server.pyi @@ -25,6 +25,7 @@ import synapse.server_notices.server_notices_manager import synapse.server_notices.server_notices_sender import synapse.state import synapse.storage +from synapse.events.builder import EventBuilderFactory class HomeServer(object): @property @@ -121,3 +122,7 @@ class HomeServer(object): pass def get_instance_id(self) -> str: pass + def get_event_builder_factory(self) -> EventBuilderFactory: + pass + def get_storage(self) -> synapse.storage.Storage: + pass diff --git a/synapse/storage/data_stores/main/events_worker.py b/synapse/storage/data_stores/main/events_worker.py index ce8be72bfe..73df6b33ba 100644 --- a/synapse/storage/data_stores/main/events_worker.py +++ b/synapse/storage/data_stores/main/events_worker.py @@ -19,7 +19,7 @@ import itertools import logging import threading from collections import namedtuple -from typing import List, Optional +from typing import List, Optional, Tuple from canonicaljson import json from constantly import NamedConstant, Names @@ -1084,7 +1084,28 @@ class EventsWorkerStore(SQLBaseStore): "get_all_new_backfill_event_rows", get_all_new_backfill_event_rows ) - def get_all_updated_current_state_deltas(self, from_token, to_token, limit): + async def get_all_updated_current_state_deltas( + self, from_token: int, to_token: int, target_row_count: int + ) -> Tuple[List[Tuple], int, bool]: + """Fetch updates from current_state_delta_stream + + Args: + from_token: The previous stream token. Updates from this stream id will + be excluded. + + to_token: The current stream token (ie the upper limit). Updates up to this + stream id will be included (modulo the 'limit' param) + + target_row_count: The number of rows to try to return. If more rows are + available, we will set 'limited' in the result. In the event of a large + batch, we may return more rows than this. + Returns: + A triplet `(updates, new_last_token, limited)`, where: + * `updates` is a list of database tuples. + * `new_last_token` is the new position in stream. + * `limited` is whether there are more updates to fetch. + """ + def get_all_updated_current_state_deltas_txn(txn): sql = """ SELECT stream_id, room_id, type, state_key, event_id @@ -1092,10 +1113,45 @@ class EventsWorkerStore(SQLBaseStore): WHERE ? < stream_id AND stream_id <= ? ORDER BY stream_id ASC LIMIT ? """ - txn.execute(sql, (from_token, to_token, limit)) + txn.execute(sql, (from_token, to_token, target_row_count)) return txn.fetchall() - return self.db.runInteraction( + def get_deltas_for_stream_id_txn(txn, stream_id): + sql = """ + SELECT stream_id, room_id, type, state_key, event_id + FROM current_state_delta_stream + WHERE stream_id = ? + """ + txn.execute(sql, [stream_id]) + return txn.fetchall() + + # we need to make sure that, for every stream id in the results, we get *all* + # the rows with that stream id. + + rows = await self.db.runInteraction( "get_all_updated_current_state_deltas", get_all_updated_current_state_deltas_txn, + ) # type: List[Tuple] + + # if we've got fewer rows than the limit, we're good + if len(rows) < target_row_count: + return rows, to_token, False + + # we hit the limit, so reduce the upper limit so that we exclude the stream id + # of the last row in the result. + assert rows[-1][0] <= to_token + to_token = rows[-1][0] - 1 + + # search backwards through the list for the point to truncate + for idx in range(len(rows) - 1, 0, -1): + if rows[idx - 1][0] <= to_token: + return rows[:idx], to_token, True + + # bother. We didn't get a full set of changes for even a single + # stream id. let's run the query again, without a row limit, but for + # just one stream id. + to_token += 1 + rows = await self.db.runInteraction( + "get_deltas_for_stream_id", get_deltas_for_stream_id_txn, to_token ) + return rows, to_token, True diff --git a/tests/replication/tcp/streams/_base.py b/tests/replication/tcp/streams/_base.py index 82f15c64e0..83e16cfe3d 100644 --- a/tests/replication/tcp/streams/_base.py +++ b/tests/replication/tcp/streams/_base.py @@ -12,10 +12,9 @@ # 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 typing import Optional -from mock import Mock +import logging +from typing import Any, Dict, List, Optional, Tuple import attr @@ -25,6 +24,7 @@ from twisted.web.http import HTTPChannel from synapse.app.generic_worker import GenericWorkerServer from synapse.http.site import SynapseRequest +from synapse.replication.slave.storage._base import BaseSlavedStore from synapse.replication.tcp.client import ReplicationDataHandler from synapse.replication.tcp.handler import ReplicationCommandHandler from synapse.replication.tcp.protocol import ClientReplicationStreamProtocol @@ -65,9 +65,7 @@ class BaseStreamTestCase(unittest.HomeserverTestCase): # databases objects are the same. self.worker_hs.get_datastore().db = hs.get_datastore().db - self.test_handler = Mock( - wraps=TestReplicationDataHandler(self.worker_hs.get_datastore()) - ) + self.test_handler = self._build_replication_data_handler() self.worker_hs.replication_data_handler = self.test_handler repl_handler = ReplicationCommandHandler(self.worker_hs) @@ -78,6 +76,9 @@ class BaseStreamTestCase(unittest.HomeserverTestCase): self._client_transport = None self._server_transport = None + def _build_replication_data_handler(self): + return TestReplicationDataHandler(self.worker_hs.get_datastore()) + def reconnect(self): if self._client_transport: self.client.close() @@ -174,22 +175,28 @@ class BaseStreamTestCase(unittest.HomeserverTestCase): class TestReplicationDataHandler(ReplicationDataHandler): """Drop-in for ReplicationDataHandler which just collects RDATA rows""" - def __init__(self, hs): - super().__init__(hs) - self.streams = set() - self._received_rdata_rows = [] + def __init__(self, store: BaseSlavedStore): + super().__init__(store) + + # streams to subscribe to: map from stream id to position + self.stream_positions = {} # type: Dict[str, int] + + # list of received (stream_name, token, row) tuples + self.received_rdata_rows = [] # type: List[Tuple[str, int, Any]] def get_streams_to_replicate(self): - positions = {s: 0 for s in self.streams} - for stream, token, _ in self._received_rdata_rows: - if stream in self.streams: - positions[stream] = max(token, positions.get(stream, 0)) - return positions + return self.stream_positions async def on_rdata(self, stream_name, token, rows): await super().on_rdata(stream_name, token, rows) for r in rows: - self._received_rdata_rows.append((stream_name, token, r)) + self.received_rdata_rows.append((stream_name, token, r)) + + if ( + stream_name in self.stream_positions + and token > self.stream_positions[stream_name] + ): + self.stream_positions[stream_name] = token @attr.s() @@ -221,7 +228,7 @@ class _PushHTTPChannel(HTTPChannel): super().__init__() self.reactor = reactor - self._pull_to_push_producer = None + self._pull_to_push_producer = None # type: Optional[_PullToPushProducer] def registerProducer(self, producer, streaming): # Convert pull producers to push producer. diff --git a/tests/replication/tcp/streams/test_events.py b/tests/replication/tcp/streams/test_events.py new file mode 100644 index 0000000000..1fa28084f9 --- /dev/null +++ b/tests/replication/tcp/streams/test_events.py @@ -0,0 +1,417 @@ +# -*- coding: utf-8 -*- +# Copyright 2019 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 typing import List, Optional + +from synapse.api.constants import EventTypes, Membership +from synapse.events import EventBase +from synapse.replication.tcp.streams._base import _STREAM_UPDATE_TARGET_ROW_COUNT +from synapse.replication.tcp.streams.events import ( + EventsStreamCurrentStateRow, + EventsStreamEventRow, + EventsStreamRow, +) +from synapse.rest import admin +from synapse.rest.client.v1 import login, room + +from tests.replication.tcp.streams._base import BaseStreamTestCase +from tests.test_utils.event_injection import inject_event, inject_member_event + + +class EventsStreamTestCase(BaseStreamTestCase): + servlets = [ + admin.register_servlets, + login.register_servlets, + room.register_servlets, + ] + + def prepare(self, reactor, clock, hs): + super().prepare(reactor, clock, hs) + self.user_id = self.register_user("u1", "pass") + self.user_tok = self.login("u1", "pass") + + self.reconnect() + self.test_handler.stream_positions["events"] = 0 + + self.room_id = self.helper.create_room_as(tok=self.user_tok) + self.test_handler.received_rdata_rows.clear() + + def test_update_function_event_row_limit(self): + """Test replication with many non-state events + + Checks that all events are correctly replicated when there are lots of + event rows to be replicated. + """ + # disconnect, so that we can stack up some changes + self.disconnect() + + # generate lots of non-state events. We inject them using inject_event + # so that they are not send out over replication until we call self.replicate(). + events = [ + self._inject_test_event() + for _ in range(_STREAM_UPDATE_TARGET_ROW_COUNT + 1) + ] + + # also one state event + state_event = self._inject_state_event() + + # tell the notifier to catch up to avoid duplicate rows. + # workaround for https://github.com/matrix-org/synapse/issues/7360 + # FIXME remove this when the above is fixed + self.replicate() + + # check we're testing what we think we are: no rows should yet have been + # received + self.assertEqual([], self.test_handler.received_rdata_rows) + + # now reconnect to pull the updates + self.reconnect() + self.replicate() + + # we should have received all the expected rows in the right order + received_rows = self.test_handler.received_rdata_rows + for event in events: + stream_name, token, row = received_rows.pop(0) + self.assertEqual("events", stream_name) + self.assertIsInstance(row, EventsStreamRow) + self.assertEqual(row.type, "ev") + self.assertIsInstance(row.data, EventsStreamEventRow) + self.assertEqual(row.data.event_id, event.event_id) + + stream_name, token, row = received_rows.pop(0) + self.assertIsInstance(row, EventsStreamRow) + self.assertIsInstance(row.data, EventsStreamEventRow) + self.assertEqual(row.data.event_id, state_event.event_id) + + stream_name, token, row = received_rows.pop(0) + self.assertEqual("events", stream_name) + self.assertIsInstance(row, EventsStreamRow) + self.assertEqual(row.type, "state") + self.assertIsInstance(row.data, EventsStreamCurrentStateRow) + self.assertEqual(row.data.event_id, state_event.event_id) + + self.assertEqual([], received_rows) + + def test_update_function_huge_state_change(self): + """Test replication with many state events + + Ensures that all events are correctly replicated when there are lots of + state change rows to be replicated. + """ + + # we want to generate lots of state changes at a single stream ID. + # + # We do this by having two branches in the DAG. On one, we have a moderator + # which that generates lots of state; on the other, we de-op the moderator, + # thus invalidating all the state. + + OTHER_USER = "@other_user:localhost" + + # have the user join + inject_member_event(self.hs, self.room_id, OTHER_USER, Membership.JOIN) + + # Update existing power levels with mod at PL50 + pls = self.helper.get_state( + self.room_id, EventTypes.PowerLevels, tok=self.user_tok + ) + pls["users"][OTHER_USER] = 50 + self.helper.send_state( + self.room_id, EventTypes.PowerLevels, pls, tok=self.user_tok, + ) + + # this is the point in the DAG where we make a fork + fork_point = self.get_success( + self.hs.get_datastore().get_latest_event_ids_in_room(self.room_id) + ) # type: List[str] + + events = [ + self._inject_state_event(sender=OTHER_USER) + for _ in range(_STREAM_UPDATE_TARGET_ROW_COUNT) + ] + + self.replicate() + # all those events and state changes should have landed + self.assertGreaterEqual( + len(self.test_handler.received_rdata_rows), 2 * len(events) + ) + + # disconnect, so that we can stack up the changes + self.disconnect() + self.test_handler.received_rdata_rows.clear() + + # a state event which doesn't get rolled back, to check that the state + # before the huge update comes through ok + state1 = self._inject_state_event() + + # roll back all the state by de-modding the user + prev_events = fork_point + pls["users"][OTHER_USER] = 0 + pl_event = inject_event( + self.hs, + prev_event_ids=prev_events, + type=EventTypes.PowerLevels, + state_key="", + sender=self.user_id, + room_id=self.room_id, + content=pls, + ) + + # one more bit of state that doesn't get rolled back + state2 = self._inject_state_event() + + # tell the notifier to catch up to avoid duplicate rows. + # workaround for https://github.com/matrix-org/synapse/issues/7360 + # FIXME remove this when the above is fixed + self.replicate() + + # check we're testing what we think we are: no rows should yet have been + # received + self.assertEqual([], self.test_handler.received_rdata_rows) + + # now reconnect to pull the updates + self.reconnect() + self.replicate() + + # now we should have received all the expected rows in the right order. + # + # we expect: + # + # - two rows for state1 + # - the PL event row, plus state rows for the PL event and each + # of the states that got reverted. + # - two rows for state2 + + received_rows = self.test_handler.received_rdata_rows + + # first check the first two rows, which should be state1 + + stream_name, token, row = received_rows.pop(0) + self.assertEqual("events", stream_name) + self.assertIsInstance(row, EventsStreamRow) + self.assertEqual(row.type, "ev") + self.assertIsInstance(row.data, EventsStreamEventRow) + self.assertEqual(row.data.event_id, state1.event_id) + + stream_name, token, row = received_rows.pop(0) + self.assertIsInstance(row, EventsStreamRow) + self.assertEqual(row.type, "state") + self.assertIsInstance(row.data, EventsStreamCurrentStateRow) + self.assertEqual(row.data.event_id, state1.event_id) + + # now the last two rows, which should be state2 + stream_name, token, row = received_rows.pop(-2) + self.assertEqual("events", stream_name) + self.assertIsInstance(row, EventsStreamRow) + self.assertEqual(row.type, "ev") + self.assertIsInstance(row.data, EventsStreamEventRow) + self.assertEqual(row.data.event_id, state2.event_id) + + stream_name, token, row = received_rows.pop(-1) + self.assertIsInstance(row, EventsStreamRow) + self.assertEqual(row.type, "state") + self.assertIsInstance(row.data, EventsStreamCurrentStateRow) + self.assertEqual(row.data.event_id, state2.event_id) + + # that should leave us with the rows for the PL event + self.assertEqual(len(received_rows), len(events) + 2) + + stream_name, token, row = received_rows.pop(0) + self.assertEqual("events", stream_name) + self.assertIsInstance(row, EventsStreamRow) + self.assertEqual(row.type, "ev") + self.assertIsInstance(row.data, EventsStreamEventRow) + self.assertEqual(row.data.event_id, pl_event.event_id) + + # the state rows are unsorted + state_rows = [] # type: List[EventsStreamCurrentStateRow] + for stream_name, token, row in received_rows: + self.assertEqual("events", stream_name) + self.assertIsInstance(row, EventsStreamRow) + self.assertEqual(row.type, "state") + self.assertIsInstance(row.data, EventsStreamCurrentStateRow) + state_rows.append(row.data) + + state_rows.sort(key=lambda r: r.state_key) + + sr = state_rows.pop(0) + self.assertEqual(sr.type, EventTypes.PowerLevels) + self.assertEqual(sr.event_id, pl_event.event_id) + for sr in state_rows: + self.assertEqual(sr.type, "test_state_event") + # "None" indicates the state has been deleted + self.assertIsNone(sr.event_id) + + def test_update_function_state_row_limit(self): + """Test replication with many state events over several stream ids. + """ + + # we want to generate lots of state changes, but for this test, we want to + # spread out the state changes over a few stream IDs. + # + # We do this by having two branches in the DAG. On one, we have four moderators, + # each of which that generates lots of state; on the other, we de-op the users, + # thus invalidating all the state. + + NUM_USERS = 4 + STATES_PER_USER = _STREAM_UPDATE_TARGET_ROW_COUNT // 4 + 1 + + user_ids = ["@user%i:localhost" % (i,) for i in range(NUM_USERS)] + + # have the users join + for u in user_ids: + inject_member_event(self.hs, self.room_id, u, Membership.JOIN) + + # Update existing power levels with mod at PL50 + pls = self.helper.get_state( + self.room_id, EventTypes.PowerLevels, tok=self.user_tok + ) + pls["users"].update({u: 50 for u in user_ids}) + self.helper.send_state( + self.room_id, EventTypes.PowerLevels, pls, tok=self.user_tok, + ) + + # this is the point in the DAG where we make a fork + fork_point = self.get_success( + self.hs.get_datastore().get_latest_event_ids_in_room(self.room_id) + ) # type: List[str] + + events = [] # type: List[EventBase] + for user in user_ids: + events.extend( + self._inject_state_event(sender=user) for _ in range(STATES_PER_USER) + ) + + self.replicate() + + # all those events and state changes should have landed + self.assertGreaterEqual( + len(self.test_handler.received_rdata_rows), 2 * len(events) + ) + + # disconnect, so that we can stack up the changes + self.disconnect() + self.test_handler.received_rdata_rows.clear() + + # now roll back all that state by de-modding the users + prev_events = fork_point + pl_events = [] + for u in user_ids: + pls["users"][u] = 0 + e = inject_event( + self.hs, + prev_event_ids=prev_events, + type=EventTypes.PowerLevels, + state_key="", + sender=self.user_id, + room_id=self.room_id, + content=pls, + ) + prev_events = [e.event_id] + pl_events.append(e) + + # tell the notifier to catch up to avoid duplicate rows. + # workaround for https://github.com/matrix-org/synapse/issues/7360 + # FIXME remove this when the above is fixed + self.replicate() + + # check we're testing what we think we are: no rows should yet have been + # received + self.assertEqual([], self.test_handler.received_rdata_rows) + + # now reconnect to pull the updates + self.reconnect() + self.replicate() + + # we should have received all the expected rows in the right order + + received_rows = self.test_handler.received_rdata_rows + self.assertGreaterEqual(len(received_rows), len(events)) + for i in range(NUM_USERS): + # for each user, we expect the PL event row, followed by state rows for + # the PL event and each of the states that got reverted. + stream_name, token, row = received_rows.pop(0) + self.assertEqual("events", stream_name) + self.assertIsInstance(row, EventsStreamRow) + self.assertEqual(row.type, "ev") + self.assertIsInstance(row.data, EventsStreamEventRow) + self.assertEqual(row.data.event_id, pl_events[i].event_id) + + # the state rows are unsorted + state_rows = [] # type: List[EventsStreamCurrentStateRow] + for j in range(STATES_PER_USER + 1): + stream_name, token, row = received_rows.pop(0) + self.assertEqual("events", stream_name) + self.assertIsInstance(row, EventsStreamRow) + self.assertEqual(row.type, "state") + self.assertIsInstance(row.data, EventsStreamCurrentStateRow) + state_rows.append(row.data) + + state_rows.sort(key=lambda r: r.state_key) + + sr = state_rows.pop(0) + self.assertEqual(sr.type, EventTypes.PowerLevels) + self.assertEqual(sr.event_id, pl_events[i].event_id) + for sr in state_rows: + self.assertEqual(sr.type, "test_state_event") + # "None" indicates the state has been deleted + self.assertIsNone(sr.event_id) + + self.assertEqual([], received_rows) + + event_count = 0 + + def _inject_test_event( + self, body: Optional[str] = None, sender: Optional[str] = None, **kwargs + ) -> EventBase: + if sender is None: + sender = self.user_id + + if body is None: + body = "event %i" % (self.event_count,) + self.event_count += 1 + + return inject_event( + self.hs, + room_id=self.room_id, + sender=sender, + type="test_event", + content={"body": body}, + **kwargs + ) + + def _inject_state_event( + self, + body: Optional[str] = None, + state_key: Optional[str] = None, + sender: Optional[str] = None, + ) -> EventBase: + if sender is None: + sender = self.user_id + + if state_key is None: + state_key = "state_%i" % (self.event_count,) + self.event_count += 1 + + if body is None: + body = "state event %s" % (state_key,) + + return inject_event( + self.hs, + room_id=self.room_id, + sender=sender, + type="test_state_event", + state_key=state_key, + content={"body": body}, + ) diff --git a/tests/replication/tcp/streams/test_receipts.py b/tests/replication/tcp/streams/test_receipts.py index a0206f7363..c122b8589c 100644 --- a/tests/replication/tcp/streams/test_receipts.py +++ b/tests/replication/tcp/streams/test_receipts.py @@ -12,6 +12,11 @@ # 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. + +# type: ignore + +from mock import Mock + from synapse.replication.tcp.streams._base import ReceiptsStream from tests.replication.tcp.streams._base import BaseStreamTestCase @@ -20,11 +25,14 @@ USER_ID = "@feeling:blue" class ReceiptsStreamTestCase(BaseStreamTestCase): + def _build_replication_data_handler(self): + return Mock(wraps=super()._build_replication_data_handler()) + def test_receipt(self): self.reconnect() # make the client subscribe to the receipts stream - self.test_handler.streams.add("receipts") + self.test_handler.stream_positions.update({"receipts": 0}) # tell the master to send a new receipt self.get_success( diff --git a/tests/replication/tcp/streams/test_typing.py b/tests/replication/tcp/streams/test_typing.py index f0ad6402ae..4d354a9db8 100644 --- a/tests/replication/tcp/streams/test_typing.py +++ b/tests/replication/tcp/streams/test_typing.py @@ -12,6 +12,8 @@ # 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 mock import Mock + from synapse.handlers.typing import RoomMember from synapse.replication.http import streams from synapse.replication.tcp.streams import TypingStream @@ -26,6 +28,9 @@ class TypingStreamTestCase(BaseStreamTestCase): streams.register_servlets, ] + def _build_replication_data_handler(self): + return Mock(wraps=super()._build_replication_data_handler()) + def test_typing(self): typing = self.hs.get_typing_handler() @@ -33,8 +38,8 @@ class TypingStreamTestCase(BaseStreamTestCase): self.reconnect() - # make the client subscribe to the receipts stream - self.test_handler.streams.add("typing") + # make the client subscribe to the typing stream + self.test_handler.stream_positions.update({"typing": 0}) typing._push_update(member=RoomMember(room_id, USER_ID), typing=True) @@ -75,6 +80,6 @@ class TypingStreamTestCase(BaseStreamTestCase): stream_name, token, rdata_rows = self.test_handler.on_rdata.call_args[0] self.assertEqual(stream_name, "typing") self.assertEqual(1, len(rdata_rows)) - row = rdata_rows[0] # type: TypingStream.TypingStreamRow + row = rdata_rows[0] self.assertEqual(room_id, row.room_id) self.assertEqual([], row.user_ids) diff --git a/tests/rest/client/v1/utils.py b/tests/rest/client/v1/utils.py index 371637618d..22d734e763 100644 --- a/tests/rest/client/v1/utils.py +++ b/tests/rest/client/v1/utils.py @@ -39,7 +39,7 @@ class RestHelper(object): resource = attr.ib() auth_user_id = attr.ib() - def create_room_as(self, room_creator, is_public=True, tok=None): + def create_room_as(self, room_creator=None, is_public=True, tok=None): temp_id = self.auth_user_id self.auth_user_id = room_creator path = "/_matrix/client/r0/createRoom" diff --git a/tests/test_utils/__init__.py b/tests/test_utils/__init__.py index a7310cf12a..7b345b03bb 100644 --- a/tests/test_utils/__init__.py +++ b/tests/test_utils/__init__.py @@ -1,5 +1,6 @@ # -*- coding: utf-8 -*- # Copyright 2019 New Vector Ltd +# Copyright 2020 The Matrix.org Foundation C.I.C # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. @@ -16,3 +17,22 @@ """ Utilities for running the unit tests """ +from typing import Awaitable, TypeVar + +TV = TypeVar("TV") + + +def get_awaitable_result(awaitable: Awaitable[TV]) -> TV: + """Get the result from an Awaitable which should have completed + + Asserts that the given awaitable has a result ready, and returns its value + """ + i = awaitable.__await__() + try: + next(i) + except StopIteration as e: + # awaitable returned a result + return e.value + + # if next didn't raise, the awaitable hasn't completed. + raise Exception("awaitable has not yet completed") diff --git a/tests/test_utils/event_injection.py b/tests/test_utils/event_injection.py new file mode 100644 index 0000000000..8f6872761a --- /dev/null +++ b/tests/test_utils/event_injection.py @@ -0,0 +1,96 @@ +# -*- coding: utf-8 -*- +# Copyright 2018 New Vector Ltd +# Copyright 2020 The Matrix.org Foundation C.I.C +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +from typing import Optional + +import synapse.server +from synapse.api.constants import EventTypes +from synapse.api.room_versions import KNOWN_ROOM_VERSIONS +from synapse.events import EventBase +from synapse.types import Collection + +from tests.test_utils import get_awaitable_result + + +""" +Utility functions for poking events into the storage of the server under test. +""" + + +def inject_member_event( + hs: synapse.server.HomeServer, + room_id: str, + sender: str, + membership: str, + target: Optional[str] = None, + extra_content: Optional[dict] = None, + **kwargs +) -> EventBase: + """Inject a membership event into a room.""" + if target is None: + target = sender + + content = {"membership": membership} + if extra_content: + content.update(extra_content) + + return inject_event( + hs, + room_id=room_id, + type=EventTypes.Member, + sender=sender, + state_key=target, + content=content, + **kwargs + ) + + +def inject_event( + hs: synapse.server.HomeServer, + room_version: Optional[str] = None, + prev_event_ids: Optional[Collection[str]] = None, + **kwargs +) -> EventBase: + """Inject a generic event into a room + + Args: + hs: the homeserver under test + room_version: the version of the room we're inserting into. + if not specified, will be looked up + prev_event_ids: prev_events for the event. If not specified, will be looked up + kwargs: fields for the event to be created + """ + test_reactor = hs.get_reactor() + + if room_version is None: + d = hs.get_datastore().get_room_version_id(kwargs["room_id"]) + test_reactor.advance(0) + room_version = get_awaitable_result(d) + + builder = hs.get_event_builder_factory().for_room_version( + KNOWN_ROOM_VERSIONS[room_version], kwargs + ) + d = hs.get_event_creation_handler().create_new_client_event( + builder, prev_event_ids=prev_event_ids + ) + test_reactor.advance(0) + event, context = get_awaitable_result(d) + + d = hs.get_storage().persistence.persist_event(event, context) + test_reactor.advance(0) + get_awaitable_result(d) + + return event diff --git a/tests/unittest.py b/tests/unittest.py index 27af5228fe..6b6f224e9c 100644 --- a/tests/unittest.py +++ b/tests/unittest.py @@ -32,7 +32,6 @@ from twisted.python.threadpool import ThreadPool from twisted.trial import unittest from synapse.api.constants import EventTypes, Membership -from synapse.api.room_versions import KNOWN_ROOM_VERSIONS from synapse.config.homeserver import HomeServerConfig from synapse.config.ratelimiting import FederationRateLimitConfig from synapse.federation.transport import server as federation_server @@ -55,6 +54,7 @@ from tests.server import ( render, setup_test_homeserver, ) +from tests.test_utils import event_injection from tests.test_utils.logging_setup import setup_logging from tests.utils import default_config, setupdb @@ -596,36 +596,14 @@ class HomeserverTestCase(TestCase): """ Inject a membership event into a room. + Deprecated: use event_injection.inject_room_member directly + Args: room: Room ID to inject the event into. user: MXID of the user to inject the membership for. membership: The membership type. """ - event_builder_factory = self.hs.get_event_builder_factory() - event_creation_handler = self.hs.get_event_creation_handler() - - room_version = self.get_success( - self.hs.get_datastore().get_room_version_id(room) - ) - - builder = event_builder_factory.for_room_version( - KNOWN_ROOM_VERSIONS[room_version], - { - "type": EventTypes.Member, - "sender": user, - "state_key": user, - "room_id": room, - "content": {"membership": membership}, - }, - ) - - event, context = self.get_success( - event_creation_handler.create_new_client_event(builder) - ) - - self.get_success( - self.hs.get_storage().persistence.persist_event(event, context) - ) + event_injection.inject_member_event(self.hs, room, user, membership) class FederatingHomeserverTestCase(HomeserverTestCase): diff --git a/tox.ini b/tox.ini index 31011d7436..2630857436 100644 --- a/tox.ini +++ b/tox.ini @@ -204,6 +204,8 @@ commands = mypy \ synapse/storage/database.py \ synapse/streams \ synapse/util/caches/stream_change_cache.py \ + tests/replication/tcp/streams \ + tests/test_utils \ tests/util/test_stream_change_cache.py # To find all folders that pass mypy you run: -- cgit 1.5.1 From 3085cde577216519d789c8160262831cb2029972 Mon Sep 17 00:00:00 2001 From: Erik Johnston Date: Fri, 1 May 2020 15:21:35 +0100 Subject: Use `stream.current_token()` and remove `stream_positions()` (#7172) We move the processing of typing and federation replication traffic into their handlers so that `Stream.current_token()` points to a valid token. This allows us to remove `get_streams_to_replicate()` and `stream_positions()`. --- changelog.d/7172.misc | 1 + synapse/app/generic_worker.py | 16 ------------ synapse/replication/slave/storage/_base.py | 15 +----------- synapse/replication/slave/storage/account_data.py | 8 ------ synapse/replication/slave/storage/deviceinbox.py | 5 ---- synapse/replication/slave/storage/devices.py | 10 -------- synapse/replication/slave/storage/events.py | 6 ----- synapse/replication/slave/storage/groups.py | 5 ---- synapse/replication/slave/storage/presence.py | 9 ------- synapse/replication/slave/storage/push_rule.py | 5 ---- synapse/replication/slave/storage/pushers.py | 5 ---- synapse/replication/slave/storage/receipts.py | 5 ---- synapse/replication/slave/storage/room.py | 5 ---- synapse/replication/tcp/client.py | 19 +------------- synapse/replication/tcp/handler.py | 10 +------- tests/replication/tcp/streams/_base.py | 30 ++++++++--------------- tests/replication/tcp/streams/test_events.py | 24 ++++++++++++------ tests/replication/tcp/streams/test_receipts.py | 3 --- tests/replication/tcp/streams/test_typing.py | 3 --- 19 files changed, 30 insertions(+), 154 deletions(-) create mode 100644 changelog.d/7172.misc (limited to 'tests/replication/tcp/streams/_base.py') diff --git a/changelog.d/7172.misc b/changelog.d/7172.misc new file mode 100644 index 0000000000..ffecdf97fe --- /dev/null +++ b/changelog.d/7172.misc @@ -0,0 +1 @@ +Use `stream.current_token()` and remove `stream_positions()`. diff --git a/synapse/app/generic_worker.py b/synapse/app/generic_worker.py index 0ace7b787d..97b9b81237 100644 --- a/synapse/app/generic_worker.py +++ b/synapse/app/generic_worker.py @@ -413,12 +413,6 @@ class GenericWorkerTyping(object): # map room IDs to sets of users currently typing self._room_typing = {} - def stream_positions(self): - # We must update this typing token from the response of the previous - # sync. In particular, the stream id may "reset" back to zero/a low - # value which we *must* use for the next replication request. - return {"typing": self._latest_room_serial} - def process_replication_rows(self, token, rows): if self._latest_room_serial > token: # The master has gone backwards. To prevent inconsistent data, just @@ -658,13 +652,6 @@ class GenericWorkerReplicationHandler(ReplicationDataHandler): ) await self.process_and_notify(stream_name, token, rows) - def get_streams_to_replicate(self): - args = super(GenericWorkerReplicationHandler, self).get_streams_to_replicate() - args.update(self.typing_handler.stream_positions()) - if self.send_handler: - args.update(self.send_handler.stream_positions()) - return args - async def process_and_notify(self, stream_name, token, rows): try: if self.send_handler: @@ -799,9 +786,6 @@ class FederationSenderHandler(object): def wake_destination(self, server: str): self.federation_sender.wake_destination(server) - def stream_positions(self): - return {"federation": self.federation_position} - async def process_replication_rows(self, stream_name, token, rows): # The federation stream contains things that we want to send out, e.g. # presence, typing, etc. diff --git a/synapse/replication/slave/storage/_base.py b/synapse/replication/slave/storage/_base.py index 751c799d94..5d7c8871a4 100644 --- a/synapse/replication/slave/storage/_base.py +++ b/synapse/replication/slave/storage/_base.py @@ -14,7 +14,7 @@ # limitations under the License. import logging -from typing import Dict, Optional +from typing import Optional import six @@ -49,19 +49,6 @@ class BaseSlavedStore(CacheInvalidationWorkerStore): self.hs = hs - def stream_positions(self) -> Dict[str, int]: - """ - Get the current positions of all the streams this store wants to subscribe to - - Returns: - map from stream name to the most recent update we have for - that stream (ie, the point we want to start replicating from) - """ - pos = {} - if self._cache_id_gen: - pos["caches"] = self._cache_id_gen.get_current_token() - return pos - def get_cache_stream_token(self): if self._cache_id_gen: return self._cache_id_gen.get_current_token() diff --git a/synapse/replication/slave/storage/account_data.py b/synapse/replication/slave/storage/account_data.py index ebe94909cb..65e54b1c71 100644 --- a/synapse/replication/slave/storage/account_data.py +++ b/synapse/replication/slave/storage/account_data.py @@ -32,14 +32,6 @@ class SlavedAccountDataStore(TagsWorkerStore, AccountDataWorkerStore, BaseSlaved def get_max_account_data_stream_id(self): return self._account_data_id_gen.get_current_token() - def stream_positions(self): - result = super(SlavedAccountDataStore, self).stream_positions() - position = self._account_data_id_gen.get_current_token() - result["user_account_data"] = position - result["room_account_data"] = position - result["tag_account_data"] = position - return result - def process_replication_rows(self, stream_name, token, rows): if stream_name == "tag_account_data": self._account_data_id_gen.advance(token) diff --git a/synapse/replication/slave/storage/deviceinbox.py b/synapse/replication/slave/storage/deviceinbox.py index 0c237c6e0f..c923751e50 100644 --- a/synapse/replication/slave/storage/deviceinbox.py +++ b/synapse/replication/slave/storage/deviceinbox.py @@ -43,11 +43,6 @@ class SlavedDeviceInboxStore(DeviceInboxWorkerStore, BaseSlavedStore): expiry_ms=30 * 60 * 1000, ) - def stream_positions(self): - result = super(SlavedDeviceInboxStore, self).stream_positions() - result["to_device"] = self._device_inbox_id_gen.get_current_token() - return result - def process_replication_rows(self, stream_name, token, rows): if stream_name == "to_device": self._device_inbox_id_gen.advance(token) diff --git a/synapse/replication/slave/storage/devices.py b/synapse/replication/slave/storage/devices.py index 23b1650e41..58fb0eaae3 100644 --- a/synapse/replication/slave/storage/devices.py +++ b/synapse/replication/slave/storage/devices.py @@ -48,16 +48,6 @@ class SlavedDeviceStore(EndToEndKeyWorkerStore, DeviceWorkerStore, BaseSlavedSto "DeviceListFederationStreamChangeCache", device_list_max ) - def stream_positions(self): - result = super(SlavedDeviceStore, self).stream_positions() - # The user signature stream uses the same stream ID generator as the - # device list stream, so set them both to the device list ID - # generator's current token. - current_token = self._device_list_id_gen.get_current_token() - result[DeviceListsStream.NAME] = current_token - result[UserSignatureStream.NAME] = current_token - return result - def process_replication_rows(self, stream_name, token, rows): if stream_name == DeviceListsStream.NAME: self._device_list_id_gen.advance(token) diff --git a/synapse/replication/slave/storage/events.py b/synapse/replication/slave/storage/events.py index e73342c657..15011259df 100644 --- a/synapse/replication/slave/storage/events.py +++ b/synapse/replication/slave/storage/events.py @@ -93,12 +93,6 @@ class SlavedEventStore( 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() - result["events"] = self._stream_id_gen.get_current_token() - result["backfill"] = -self._backfill_id_gen.get_current_token() - return result - def process_replication_rows(self, stream_name, token, rows): if stream_name == "events": self._stream_id_gen.advance(token) diff --git a/synapse/replication/slave/storage/groups.py b/synapse/replication/slave/storage/groups.py index 2d4fd08cf5..01bcf0e882 100644 --- a/synapse/replication/slave/storage/groups.py +++ b/synapse/replication/slave/storage/groups.py @@ -37,11 +37,6 @@ class SlavedGroupServerStore(GroupServerWorkerStore, BaseSlavedStore): def get_group_stream_token(self): return self._group_updates_id_gen.get_current_token() - def stream_positions(self): - result = super(SlavedGroupServerStore, self).stream_positions() - result["groups"] = self._group_updates_id_gen.get_current_token() - return result - def process_replication_rows(self, stream_name, token, rows): if stream_name == "groups": self._group_updates_id_gen.advance(token) diff --git a/synapse/replication/slave/storage/presence.py b/synapse/replication/slave/storage/presence.py index ad8f0c15a9..fae3125072 100644 --- a/synapse/replication/slave/storage/presence.py +++ b/synapse/replication/slave/storage/presence.py @@ -41,15 +41,6 @@ class SlavedPresenceStore(BaseSlavedStore): def get_current_presence_token(self): return self._presence_id_gen.get_current_token() - def stream_positions(self): - result = super(SlavedPresenceStore, self).stream_positions() - - if self.hs.config.use_presence: - position = self._presence_id_gen.get_current_token() - result["presence"] = position - - return result - def process_replication_rows(self, stream_name, token, rows): if stream_name == "presence": self._presence_id_gen.advance(token) diff --git a/synapse/replication/slave/storage/push_rule.py b/synapse/replication/slave/storage/push_rule.py index eebd5a1fb6..6138796da4 100644 --- a/synapse/replication/slave/storage/push_rule.py +++ b/synapse/replication/slave/storage/push_rule.py @@ -37,11 +37,6 @@ class SlavedPushRuleStore(SlavedEventStore, PushRulesWorkerStore): 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() - return result - def process_replication_rows(self, stream_name, token, rows): if stream_name == "push_rules": self._push_rules_stream_id_gen.advance(token) diff --git a/synapse/replication/slave/storage/pushers.py b/synapse/replication/slave/storage/pushers.py index bce8a3d115..67be337945 100644 --- a/synapse/replication/slave/storage/pushers.py +++ b/synapse/replication/slave/storage/pushers.py @@ -28,11 +28,6 @@ class SlavedPusherStore(PusherWorkerStore, BaseSlavedStore): db_conn, "pushers", "id", extra_tables=[("deleted_pushers", "stream_id")] ) - def stream_positions(self): - result = super(SlavedPusherStore, self).stream_positions() - result["pushers"] = self._pushers_id_gen.get_current_token() - return result - def get_pushers_stream_token(self): return self._pushers_id_gen.get_current_token() diff --git a/synapse/replication/slave/storage/receipts.py b/synapse/replication/slave/storage/receipts.py index d40dc6e1f5..993432edcb 100644 --- a/synapse/replication/slave/storage/receipts.py +++ b/synapse/replication/slave/storage/receipts.py @@ -42,11 +42,6 @@ class SlavedReceiptsStore(ReceiptsWorkerStore, BaseSlavedStore): 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() - result["receipts"] = self._receipts_id_gen.get_current_token() - return result - def invalidate_caches_for_receipt(self, room_id, receipt_type, user_id): self.get_receipts_for_user.invalidate((user_id, receipt_type)) self._get_linearized_receipts_for_room.invalidate_many((room_id,)) diff --git a/synapse/replication/slave/storage/room.py b/synapse/replication/slave/storage/room.py index 3a20f45316..10dda8708f 100644 --- a/synapse/replication/slave/storage/room.py +++ b/synapse/replication/slave/storage/room.py @@ -30,11 +30,6 @@ class RoomStore(RoomWorkerStore, BaseSlavedStore): 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() - result["public_rooms"] = self._public_room_id_gen.get_current_token() - return result - def process_replication_rows(self, stream_name, token, rows): if stream_name == "public_rooms": self._public_room_id_gen.advance(token) diff --git a/synapse/replication/tcp/client.py b/synapse/replication/tcp/client.py index 2d07b8b2d0..5c28fd4ac3 100644 --- a/synapse/replication/tcp/client.py +++ b/synapse/replication/tcp/client.py @@ -16,7 +16,7 @@ """ import logging -from typing import TYPE_CHECKING, Dict +from typing import TYPE_CHECKING from twisted.internet.protocol import ReconnectingClientFactory @@ -100,23 +100,6 @@ class ReplicationDataHandler: """ self.store.process_replication_rows(stream_name, token, rows) - def get_streams_to_replicate(self) -> Dict[str, int]: - """Called when a new connection has been established and we need to - subscribe to streams. - - Returns: - map from stream name to the most recent update we have for - that stream (ie, the point we want to start replicating from) - """ - args = self.store.stream_positions() - user_account_data = args.pop("user_account_data", None) - room_account_data = args.pop("room_account_data", None) - if user_account_data: - args["account_data"] = user_account_data - elif room_account_data: - args["account_data"] = room_account_data - return args - async def on_position(self, stream_name: str, token: int): self.store.process_replication_rows(stream_name, token, []) diff --git a/synapse/replication/tcp/handler.py b/synapse/replication/tcp/handler.py index 6f7054d5af..d72f3d0cf9 100644 --- a/synapse/replication/tcp/handler.py +++ b/synapse/replication/tcp/handler.py @@ -314,15 +314,7 @@ class ReplicationCommandHandler: self._pending_batches.pop(cmd.stream_name, []) # Find where we previously streamed up to. - current_token = self._replication_data_handler.get_streams_to_replicate().get( - cmd.stream_name - ) - if current_token is None: - logger.warning( - "Got POSITION for stream we're not subscribed to: %s", - cmd.stream_name, - ) - return + current_token = stream.current_token() # If the position token matches our current token then we're up to # date and there's nothing to do. Otherwise, fetch all updates diff --git a/tests/replication/tcp/streams/_base.py b/tests/replication/tcp/streams/_base.py index 83e16cfe3d..8c104f8d1d 100644 --- a/tests/replication/tcp/streams/_base.py +++ b/tests/replication/tcp/streams/_base.py @@ -14,7 +14,7 @@ # limitations under the License. import logging -from typing import Any, Dict, List, Optional, Tuple +from typing import Any, List, Optional, Tuple import attr @@ -22,13 +22,15 @@ from twisted.internet.interfaces import IConsumer, IPullProducer, IReactorTime from twisted.internet.task import LoopingCall from twisted.web.http import HTTPChannel -from synapse.app.generic_worker import GenericWorkerServer +from synapse.app.generic_worker import ( + GenericWorkerReplicationHandler, + GenericWorkerServer, +) from synapse.http.site import SynapseRequest -from synapse.replication.slave.storage._base import BaseSlavedStore -from synapse.replication.tcp.client import ReplicationDataHandler from synapse.replication.tcp.handler import ReplicationCommandHandler from synapse.replication.tcp.protocol import ClientReplicationStreamProtocol from synapse.replication.tcp.resource import ReplicationStreamProtocolFactory +from synapse.server import HomeServer from synapse.util import Clock from tests import unittest @@ -77,7 +79,7 @@ class BaseStreamTestCase(unittest.HomeserverTestCase): self._server_transport = None def _build_replication_data_handler(self): - return TestReplicationDataHandler(self.worker_hs.get_datastore()) + return TestReplicationDataHandler(self.worker_hs) def reconnect(self): if self._client_transport: @@ -172,32 +174,20 @@ class BaseStreamTestCase(unittest.HomeserverTestCase): self.assertEqual(request.method, b"GET") -class TestReplicationDataHandler(ReplicationDataHandler): +class TestReplicationDataHandler(GenericWorkerReplicationHandler): """Drop-in for ReplicationDataHandler which just collects RDATA rows""" - def __init__(self, store: BaseSlavedStore): - super().__init__(store) - - # streams to subscribe to: map from stream id to position - self.stream_positions = {} # type: Dict[str, int] + def __init__(self, hs: HomeServer): + super().__init__(hs) # list of received (stream_name, token, row) tuples self.received_rdata_rows = [] # type: List[Tuple[str, int, Any]] - def get_streams_to_replicate(self): - return self.stream_positions - async def on_rdata(self, stream_name, token, rows): await super().on_rdata(stream_name, token, rows) for r in rows: self.received_rdata_rows.append((stream_name, token, r)) - if ( - stream_name in self.stream_positions - and token > self.stream_positions[stream_name] - ): - self.stream_positions[stream_name] = token - @attr.s() class OneShotRequestFactory: diff --git a/tests/replication/tcp/streams/test_events.py b/tests/replication/tcp/streams/test_events.py index 1fa28084f9..8bd67bb9f1 100644 --- a/tests/replication/tcp/streams/test_events.py +++ b/tests/replication/tcp/streams/test_events.py @@ -43,7 +43,6 @@ class EventsStreamTestCase(BaseStreamTestCase): self.user_tok = self.login("u1", "pass") self.reconnect() - self.test_handler.stream_positions["events"] = 0 self.room_id = self.helper.create_room_as(tok=self.user_tok) self.test_handler.received_rdata_rows.clear() @@ -80,8 +79,12 @@ class EventsStreamTestCase(BaseStreamTestCase): self.reconnect() self.replicate() - # we should have received all the expected rows in the right order - received_rows = self.test_handler.received_rdata_rows + # we should have received all the expected rows in the right order (as + # well as various cache invalidation updates which we ignore) + received_rows = [ + row for row in self.test_handler.received_rdata_rows if row[0] == "events" + ] + for event in events: stream_name, token, row = received_rows.pop(0) self.assertEqual("events", stream_name) @@ -184,7 +187,8 @@ class EventsStreamTestCase(BaseStreamTestCase): self.reconnect() self.replicate() - # now we should have received all the expected rows in the right order. + # we should have received all the expected rows in the right order (as + # well as various cache invalidation updates which we ignore) # # we expect: # @@ -193,7 +197,9 @@ class EventsStreamTestCase(BaseStreamTestCase): # of the states that got reverted. # - two rows for state2 - received_rows = self.test_handler.received_rdata_rows + received_rows = [ + row for row in self.test_handler.received_rdata_rows if row[0] == "events" + ] # first check the first two rows, which should be state1 @@ -334,9 +340,11 @@ class EventsStreamTestCase(BaseStreamTestCase): self.reconnect() self.replicate() - # we should have received all the expected rows in the right order - - received_rows = self.test_handler.received_rdata_rows + # we should have received all the expected rows in the right order (as + # well as various cache invalidation updates which we ignore) + received_rows = [ + row for row in self.test_handler.received_rdata_rows if row[0] == "events" + ] self.assertGreaterEqual(len(received_rows), len(events)) for i in range(NUM_USERS): # for each user, we expect the PL event row, followed by state rows for diff --git a/tests/replication/tcp/streams/test_receipts.py b/tests/replication/tcp/streams/test_receipts.py index c122b8589c..df332ee679 100644 --- a/tests/replication/tcp/streams/test_receipts.py +++ b/tests/replication/tcp/streams/test_receipts.py @@ -31,9 +31,6 @@ class ReceiptsStreamTestCase(BaseStreamTestCase): def test_receipt(self): self.reconnect() - # make the client subscribe to the receipts stream - self.test_handler.stream_positions.update({"receipts": 0}) - # tell the master to send a new receipt self.get_success( self.hs.get_datastore().insert_receipt( diff --git a/tests/replication/tcp/streams/test_typing.py b/tests/replication/tcp/streams/test_typing.py index 4d354a9db8..e8d17ca68a 100644 --- a/tests/replication/tcp/streams/test_typing.py +++ b/tests/replication/tcp/streams/test_typing.py @@ -38,9 +38,6 @@ class TypingStreamTestCase(BaseStreamTestCase): self.reconnect() - # make the client subscribe to the typing stream - self.test_handler.stream_positions.update({"typing": 0}) - typing._push_update(member=RoomMember(room_id, USER_ID), typing=True) self.reactor.advance(0) -- cgit 1.5.1 From 0e719f23981b8294df66ba7f38b8c7cc99fad228 Mon Sep 17 00:00:00 2001 From: Erik Johnston Date: Fri, 1 May 2020 17:19:56 +0100 Subject: Thread through instance name to replication client. (#7369) For in memory streams when fetching updates on workers we need to query the source of the stream, which currently is hard coded to be master. This PR threads through the source instance we received via `POSITION` through to the update function in each stream, which can then be passed to the replication client for in memory streams. --- changelog.d/7369.misc | 1 + synapse/app/generic_worker.py | 10 +++--- synapse/replication/http/_base.py | 19 +++++++++- synapse/replication/http/streams.py | 4 ++- synapse/replication/tcp/client.py | 12 ++++--- synapse/replication/tcp/handler.py | 20 ++++++++--- synapse/replication/tcp/streams/_base.py | 50 +++++++++++++++++++------- synapse/replication/tcp/streams/events.py | 10 ++++-- synapse/replication/tcp/streams/federation.py | 4 +-- tests/replication/tcp/streams/_base.py | 4 +-- tests/replication/tcp/streams/test_receipts.py | 4 +-- tests/replication/tcp/streams/test_typing.py | 4 +-- 12 files changed, 101 insertions(+), 41 deletions(-) create mode 100644 changelog.d/7369.misc (limited to 'tests/replication/tcp/streams/_base.py') diff --git a/changelog.d/7369.misc b/changelog.d/7369.misc new file mode 100644 index 0000000000..060b09c888 --- /dev/null +++ b/changelog.d/7369.misc @@ -0,0 +1 @@ +Thread through instance name to replication client. diff --git a/synapse/app/generic_worker.py b/synapse/app/generic_worker.py index 97b9b81237..667ad20428 100644 --- a/synapse/app/generic_worker.py +++ b/synapse/app/generic_worker.py @@ -646,13 +646,11 @@ class GenericWorkerReplicationHandler(ReplicationDataHandler): else: self.send_handler = None - async def on_rdata(self, stream_name, token, rows): - await super(GenericWorkerReplicationHandler, self).on_rdata( - stream_name, token, rows - ) - await self.process_and_notify(stream_name, token, rows) + async def on_rdata(self, stream_name, instance_name, token, rows): + await super().on_rdata(stream_name, instance_name, token, rows) + await self._process_and_notify(stream_name, instance_name, token, rows) - async def process_and_notify(self, stream_name, token, rows): + async def _process_and_notify(self, stream_name, instance_name, token, rows): try: if self.send_handler: await self.send_handler.process_replication_rows( diff --git a/synapse/replication/http/_base.py b/synapse/replication/http/_base.py index 1be1ccbdf3..f88c80ae84 100644 --- a/synapse/replication/http/_base.py +++ b/synapse/replication/http/_base.py @@ -16,6 +16,7 @@ import abc import logging import re +from inspect import signature from typing import Dict, List, Tuple from six import raise_from @@ -60,6 +61,8 @@ class ReplicationEndpoint(object): must call `register` to register the path with the HTTP server. Requests can be sent by calling the client returned by `make_client`. + Requests are sent to master process by default, but can be sent to other + named processes by specifying an `instance_name` keyword argument. Attributes: NAME (str): A name for the endpoint, added to the path as well as used @@ -91,6 +94,16 @@ class ReplicationEndpoint(object): hs, "repl." + self.NAME, timeout_ms=30 * 60 * 1000 ) + # We reserve `instance_name` as a parameter to sending requests, so we + # assert here that sub classes don't try and use the name. + assert ( + "instance_name" not in self.PATH_ARGS + ), "`instance_name` is a reserved paramater name" + assert ( + "instance_name" + not in signature(self.__class__._serialize_payload).parameters + ), "`instance_name` is a reserved paramater name" + assert self.METHOD in ("PUT", "POST", "GET") @abc.abstractmethod @@ -135,7 +148,11 @@ class ReplicationEndpoint(object): @trace(opname="outgoing_replication_request") @defer.inlineCallbacks - def send_request(**kwargs): + def send_request(instance_name="master", **kwargs): + # Currently we only support sending requests to master process. + if instance_name != "master": + raise Exception("Unknown instance") + data = yield cls._serialize_payload(**kwargs) url_args = [ diff --git a/synapse/replication/http/streams.py b/synapse/replication/http/streams.py index f35cebc710..0459f582bf 100644 --- a/synapse/replication/http/streams.py +++ b/synapse/replication/http/streams.py @@ -50,6 +50,8 @@ class ReplicationGetStreamUpdates(ReplicationEndpoint): def __init__(self, hs): super().__init__(hs) + self._instance_name = hs.get_instance_name() + # We pull the streams from the replication steamer (if we try and make # them ourselves we end up in an import loop). self.streams = hs.get_replication_streamer().get_streams() @@ -67,7 +69,7 @@ class ReplicationGetStreamUpdates(ReplicationEndpoint): upto_token = parse_integer(request, "upto_token", required=True) updates, upto_token, limited = await stream.get_updates_since( - from_token, upto_token + self._instance_name, from_token, upto_token ) return ( diff --git a/synapse/replication/tcp/client.py b/synapse/replication/tcp/client.py index 5c28fd4ac3..3bbf3c3569 100644 --- a/synapse/replication/tcp/client.py +++ b/synapse/replication/tcp/client.py @@ -86,17 +86,19 @@ class ReplicationDataHandler: def __init__(self, store: BaseSlavedStore): self.store = store - async def on_rdata(self, stream_name: str, token: int, rows: list): + async def on_rdata( + self, stream_name: str, instance_name: str, token: int, rows: list + ): """Called to handle a batch of replication data with a given stream token. By default this just pokes the slave store. Can be overridden in subclasses to handle more. Args: - stream_name (str): name of the replication stream for this batch of rows - token (int): stream token for this batch of rows - rows (list): a list of Stream.ROW_TYPE objects as returned by - Stream.parse_row. + stream_name: name of the replication stream for this batch of rows + instance_name: the instance that wrote the rows. + token: stream token for this batch of rows + rows: a list of Stream.ROW_TYPE objects as returned by Stream.parse_row. """ self.store.process_replication_rows(stream_name, token, rows) diff --git a/synapse/replication/tcp/handler.py b/synapse/replication/tcp/handler.py index d72f3d0cf9..2d1d119c7c 100644 --- a/synapse/replication/tcp/handler.py +++ b/synapse/replication/tcp/handler.py @@ -278,19 +278,24 @@ class ReplicationCommandHandler: # Check if this is the last of a batch of updates rows = self._pending_batches.pop(stream_name, []) rows.append(row) - await self.on_rdata(stream_name, cmd.token, rows) + await self.on_rdata(stream_name, cmd.instance_name, cmd.token, rows) - async def on_rdata(self, stream_name: str, token: int, rows: list): + async def on_rdata( + self, stream_name: str, instance_name: str, token: int, rows: list + ): """Called to handle a batch of replication data with a given stream token. Args: stream_name: name of the replication stream for this batch of rows + instance_name: the instance that wrote the rows. token: stream token for this batch of rows rows: a list of Stream.ROW_TYPE objects as returned by Stream.parse_row. """ logger.debug("Received rdata %s -> %s", stream_name, token) - await self._replication_data_handler.on_rdata(stream_name, token, rows) + await self._replication_data_handler.on_rdata( + stream_name, instance_name, token, rows + ) async def on_POSITION(self, conn: AbstractConnection, cmd: PositionCommand): if cmd.instance_name == self._instance_name: @@ -325,7 +330,9 @@ class ReplicationCommandHandler: updates, current_token, missing_updates, - ) = await stream.get_updates_since(current_token, cmd.token) + ) = await stream.get_updates_since( + cmd.instance_name, current_token, cmd.token + ) # TODO: add some tests for this @@ -334,7 +341,10 @@ class ReplicationCommandHandler: for token, rows in _batch_updates(updates): await self.on_rdata( - cmd.stream_name, token, [stream.parse_row(row) for row in rows], + cmd.stream_name, + cmd.instance_name, + token, + [stream.parse_row(row) for row in rows], ) # We've now caught up to position sent to us, notify handler. diff --git a/synapse/replication/tcp/streams/_base.py b/synapse/replication/tcp/streams/_base.py index 4af1afd119..b0f87c365b 100644 --- a/synapse/replication/tcp/streams/_base.py +++ b/synapse/replication/tcp/streams/_base.py @@ -16,7 +16,7 @@ import logging from collections import namedtuple -from typing import Any, Awaitable, Callable, Iterable, List, Optional, Tuple +from typing import Any, Awaitable, Callable, List, Optional, Tuple import attr @@ -53,6 +53,7 @@ StreamUpdateResult = Tuple[List[Tuple[Token, StreamRow]], Token, bool] # # The arguments are: # +# * instance_name: the writer of the stream # * from_token: the previous stream token: the starting point for fetching the # updates # * to_token: the new stream token: the point to get updates up to @@ -62,7 +63,7 @@ StreamUpdateResult = Tuple[List[Tuple[Token, StreamRow]], Token, bool] # If there are more updates available, it should set `limited` in the result, and # it will be called again to get the next batch. # -UpdateFunction = Callable[[Token, Token, int], Awaitable[StreamUpdateResult]] +UpdateFunction = Callable[[str, Token, Token, int], Awaitable[StreamUpdateResult]] class Stream(object): @@ -93,6 +94,7 @@ class Stream(object): def __init__( self, + local_instance_name: str, current_token_function: Callable[[], Token], update_function: UpdateFunction, ): @@ -108,9 +110,11 @@ class Stream(object): stream tokens. See the UpdateFunction type definition for more info. Args: + local_instance_name: The instance name of the current process current_token_function: callback to get the current token, as above update_function: callback go get stream updates, as above """ + self.local_instance_name = local_instance_name self.current_token = current_token_function self.update_function = update_function @@ -135,14 +139,14 @@ class Stream(object): """ current_token = self.current_token() updates, current_token, limited = await self.get_updates_since( - self.last_token, current_token + self.local_instance_name, self.last_token, current_token ) self.last_token = current_token return updates, current_token, limited async def get_updates_since( - self, from_token: Token, upto_token: Token + self, instance_name: str, from_token: Token, upto_token: Token ) -> StreamUpdateResult: """Like get_updates except allows specifying from when we should stream updates @@ -160,19 +164,19 @@ class Stream(object): return [], upto_token, False updates, upto_token, limited = await self.update_function( - from_token, upto_token, _STREAM_UPDATE_TARGET_ROW_COUNT, + instance_name, from_token, upto_token, _STREAM_UPDATE_TARGET_ROW_COUNT, ) return updates, upto_token, limited def db_query_to_update_function( - query_function: Callable[[Token, Token, int], Awaitable[Iterable[tuple]]] + query_function: Callable[[Token, Token, int], Awaitable[List[tuple]]] ) -> UpdateFunction: """Wraps a db query function which returns a list of rows to make it suitable for use as an `update_function` for the Stream class """ - async def update_function(from_token, upto_token, limit): + async def update_function(instance_name, from_token, upto_token, limit): rows = await query_function(from_token, upto_token, limit) updates = [(row[0], row[1:]) for row in rows] limited = False @@ -193,10 +197,13 @@ def make_http_update_function(hs, stream_name: str) -> UpdateFunction: client = ReplicationGetStreamUpdates.make_client(hs) async def update_function( - from_token: int, upto_token: int, limit: int + instance_name: str, from_token: int, upto_token: int, limit: int ) -> StreamUpdateResult: result = await client( - stream_name=stream_name, from_token=from_token, upto_token=upto_token, + instance_name=instance_name, + stream_name=stream_name, + from_token=from_token, + upto_token=upto_token, ) return result["updates"], result["upto_token"], result["limited"] @@ -226,6 +233,7 @@ class BackfillStream(Stream): def __init__(self, hs): store = hs.get_datastore() super().__init__( + hs.get_instance_name(), store.get_current_backfill_token, db_query_to_update_function(store.get_all_new_backfill_event_rows), ) @@ -261,7 +269,9 @@ class PresenceStream(Stream): # Query master process update_function = make_http_update_function(hs, self.NAME) - super().__init__(store.get_current_presence_token, update_function) + super().__init__( + hs.get_instance_name(), store.get_current_presence_token, update_function + ) class TypingStream(Stream): @@ -284,7 +294,9 @@ class TypingStream(Stream): # Query master process update_function = make_http_update_function(hs, self.NAME) - super().__init__(typing_handler.get_current_token, update_function) + super().__init__( + hs.get_instance_name(), typing_handler.get_current_token, update_function + ) class ReceiptsStream(Stream): @@ -305,6 +317,7 @@ class ReceiptsStream(Stream): def __init__(self, hs): store = hs.get_datastore() super().__init__( + hs.get_instance_name(), store.get_max_receipt_stream_id, db_query_to_update_function(store.get_all_updated_receipts), ) @@ -322,14 +335,16 @@ class PushRulesStream(Stream): def __init__(self, hs): self.store = hs.get_datastore() super(PushRulesStream, self).__init__( - self._current_token, self._update_function + hs.get_instance_name(), self._current_token, self._update_function ) def _current_token(self) -> int: push_rules_token, _ = self.store.get_push_rules_stream_token() return push_rules_token - async def _update_function(self, from_token: Token, to_token: Token, limit: int): + async def _update_function( + self, instance_name: str, from_token: Token, to_token: Token, limit: int + ): rows = await self.store.get_all_push_rule_updates(from_token, to_token, limit) limited = False @@ -356,6 +371,7 @@ class PushersStream(Stream): store = hs.get_datastore() super().__init__( + hs.get_instance_name(), store.get_pushers_stream_token, db_query_to_update_function(store.get_all_updated_pushers_rows), ) @@ -387,6 +403,7 @@ class CachesStream(Stream): def __init__(self, hs): store = hs.get_datastore() super().__init__( + hs.get_instance_name(), store.get_cache_stream_token, db_query_to_update_function(store.get_all_updated_caches), ) @@ -412,6 +429,7 @@ class PublicRoomsStream(Stream): def __init__(self, hs): store = hs.get_datastore() super().__init__( + hs.get_instance_name(), store.get_current_public_room_stream_id, db_query_to_update_function(store.get_all_new_public_rooms), ) @@ -432,6 +450,7 @@ class DeviceListsStream(Stream): def __init__(self, hs): store = hs.get_datastore() super().__init__( + hs.get_instance_name(), store.get_device_stream_token, db_query_to_update_function(store.get_all_device_list_changes_for_remotes), ) @@ -449,6 +468,7 @@ class ToDeviceStream(Stream): def __init__(self, hs): store = hs.get_datastore() super().__init__( + hs.get_instance_name(), store.get_to_device_stream_token, db_query_to_update_function(store.get_all_new_device_messages), ) @@ -468,6 +488,7 @@ class TagAccountDataStream(Stream): def __init__(self, hs): store = hs.get_datastore() super().__init__( + hs.get_instance_name(), store.get_max_account_data_stream_id, db_query_to_update_function(store.get_all_updated_tags), ) @@ -487,6 +508,7 @@ class AccountDataStream(Stream): def __init__(self, hs): self.store = hs.get_datastore() super().__init__( + hs.get_instance_name(), self.store.get_max_account_data_stream_id, db_query_to_update_function(self._update_function), ) @@ -517,6 +539,7 @@ class GroupServerStream(Stream): def __init__(self, hs): store = hs.get_datastore() super().__init__( + hs.get_instance_name(), store.get_group_stream_token, db_query_to_update_function(store.get_all_groups_changes), ) @@ -534,6 +557,7 @@ class UserSignatureStream(Stream): def __init__(self, hs): store = hs.get_datastore() super().__init__( + hs.get_instance_name(), store.get_device_stream_token, db_query_to_update_function( store.get_all_user_signature_changes_for_remotes diff --git a/synapse/replication/tcp/streams/events.py b/synapse/replication/tcp/streams/events.py index 52df81b1bd..890e75d827 100644 --- a/synapse/replication/tcp/streams/events.py +++ b/synapse/replication/tcp/streams/events.py @@ -118,11 +118,17 @@ class EventsStream(Stream): def __init__(self, hs): self._store = hs.get_datastore() super().__init__( - self._store.get_current_events_token, self._update_function, + hs.get_instance_name(), + self._store.get_current_events_token, + self._update_function, ) async def _update_function( - self, from_token: Token, current_token: Token, target_row_count: int + self, + instance_name: str, + from_token: Token, + current_token: Token, + target_row_count: int, ) -> StreamUpdateResult: # the events stream merges together three separate sources: diff --git a/synapse/replication/tcp/streams/federation.py b/synapse/replication/tcp/streams/federation.py index 75133d7e40..e8bd52e389 100644 --- a/synapse/replication/tcp/streams/federation.py +++ b/synapse/replication/tcp/streams/federation.py @@ -48,8 +48,8 @@ class FederationStream(Stream): current_token = lambda: 0 update_function = self._stub_update_function - super().__init__(current_token, update_function) + super().__init__(hs.get_instance_name(), current_token, update_function) @staticmethod - async def _stub_update_function(from_token, upto_token, limit): + async def _stub_update_function(instance_name, from_token, upto_token, limit): return [], upto_token, False diff --git a/tests/replication/tcp/streams/_base.py b/tests/replication/tcp/streams/_base.py index 8c104f8d1d..7b56d2028d 100644 --- a/tests/replication/tcp/streams/_base.py +++ b/tests/replication/tcp/streams/_base.py @@ -183,8 +183,8 @@ class TestReplicationDataHandler(GenericWorkerReplicationHandler): # list of received (stream_name, token, row) tuples self.received_rdata_rows = [] # type: List[Tuple[str, int, Any]] - async def on_rdata(self, stream_name, token, rows): - await super().on_rdata(stream_name, token, rows) + async def on_rdata(self, stream_name, instance_name, token, rows): + await super().on_rdata(stream_name, instance_name, token, rows) for r in rows: self.received_rdata_rows.append((stream_name, token, r)) diff --git a/tests/replication/tcp/streams/test_receipts.py b/tests/replication/tcp/streams/test_receipts.py index df332ee679..5853314fd4 100644 --- a/tests/replication/tcp/streams/test_receipts.py +++ b/tests/replication/tcp/streams/test_receipts.py @@ -41,7 +41,7 @@ class ReceiptsStreamTestCase(BaseStreamTestCase): # there should be one RDATA command self.test_handler.on_rdata.assert_called_once() - stream_name, token, rdata_rows = self.test_handler.on_rdata.call_args[0] + stream_name, _, token, rdata_rows = self.test_handler.on_rdata.call_args[0] self.assertEqual(stream_name, "receipts") self.assertEqual(1, len(rdata_rows)) row = rdata_rows[0] # type: ReceiptsStream.ReceiptsStreamRow @@ -71,7 +71,7 @@ class ReceiptsStreamTestCase(BaseStreamTestCase): # We should now have caught up and get the missing data self.test_handler.on_rdata.assert_called_once() - stream_name, token, rdata_rows = self.test_handler.on_rdata.call_args[0] + stream_name, _, token, rdata_rows = self.test_handler.on_rdata.call_args[0] self.assertEqual(stream_name, "receipts") self.assertEqual(token, 3) self.assertEqual(1, len(rdata_rows)) diff --git a/tests/replication/tcp/streams/test_typing.py b/tests/replication/tcp/streams/test_typing.py index e8d17ca68a..d25a7b194e 100644 --- a/tests/replication/tcp/streams/test_typing.py +++ b/tests/replication/tcp/streams/test_typing.py @@ -47,7 +47,7 @@ class TypingStreamTestCase(BaseStreamTestCase): self.assert_request_is_get_repl_stream_updates(request, "typing") self.test_handler.on_rdata.assert_called_once() - stream_name, token, rdata_rows = self.test_handler.on_rdata.call_args[0] + stream_name, _, token, rdata_rows = self.test_handler.on_rdata.call_args[0] self.assertEqual(stream_name, "typing") self.assertEqual(1, len(rdata_rows)) row = rdata_rows[0] # type: TypingStream.TypingStreamRow @@ -74,7 +74,7 @@ class TypingStreamTestCase(BaseStreamTestCase): self.assertEqual(int(request.args[b"from_token"][0]), token) self.test_handler.on_rdata.assert_called_once() - stream_name, token, rdata_rows = self.test_handler.on_rdata.call_args[0] + stream_name, _, token, rdata_rows = self.test_handler.on_rdata.call_args[0] self.assertEqual(stream_name, "typing") self.assertEqual(1, len(rdata_rows)) row = rdata_rows[0] -- cgit 1.5.1