From d78265af0c77b7d75bc5fd013515b53b01a2ac23 Mon Sep 17 00:00:00 2001 From: Richard van der Hoff Date: Tue, 5 May 2020 18:53:38 +0100 Subject: Wait to subscribe before sending REPLICATE --- synapse/replication/tcp/handler.py | 3 ++- synapse/replication/tcp/redis.py | 52 ++++++++++++++++++++++++-------------- 2 files changed, 35 insertions(+), 20 deletions(-) (limited to 'synapse/replication/tcp') diff --git a/synapse/replication/tcp/handler.py b/synapse/replication/tcp/handler.py index 6f7054d5af..0e8a38fd80 100644 --- a/synapse/replication/tcp/handler.py +++ b/synapse/replication/tcp/handler.py @@ -99,7 +99,8 @@ class ReplicationCommandHandler: # The factory used to create connections. self._factory = None # type: Optional[ReconnectingClientFactory] - # The currently connected connections. + # The currently connected connections. (The list of places we need to send + # outgoing replication commands to.) self._connections = [] # type: List[AbstractConnection] LaterGauge( diff --git a/synapse/replication/tcp/redis.py b/synapse/replication/tcp/redis.py index 617e860f95..61dbf4ddbe 100644 --- a/synapse/replication/tcp/redis.py +++ b/synapse/replication/tcp/redis.py @@ -18,7 +18,7 @@ from typing import TYPE_CHECKING import txredisapi -from synapse.logging.context import PreserveLoggingContext +from synapse.logging.context import make_deferred_yieldable from synapse.metrics.background_process_metrics import run_as_background_process from synapse.replication.tcp.commands import ( Command, @@ -41,8 +41,14 @@ logger = logging.getLogger(__name__) class RedisSubscriber(txredisapi.SubscriberProtocol, AbstractConnection): """Connection to redis subscribed to replication stream. - Parses incoming messages from redis into replication commands, and passes - them to `ReplicationCommandHandler` + This class fulfils two functions: + + (a) it implements the twisted Protocol API, where it handles the SUBSCRIBEd redis + connection, parsing *incoming* messages into replication commands, and passing them + to `ReplicationCommandHandler` + + (b) it implements the AbstractConnection API, where it sends *outgoing* commands + onto outbound_redis_connection. Due to the vagaries of `txredisapi` we don't want to have a custom constructor, so instead we expect the defined attributes below to be set @@ -50,8 +56,8 @@ class RedisSubscriber(txredisapi.SubscriberProtocol, AbstractConnection): Attributes: handler: The command handler to handle incoming commands. - stream_name: The *redis* stream name to subscribe to (not anything to - do with Synapse replication streams). + stream_name: The *redis* stream name to subscribe to and publish from + (not anything to do with Synapse replication streams). outbound_redis_connection: The connection to redis to use to send commands. """ @@ -61,12 +67,22 @@ class RedisSubscriber(txredisapi.SubscriberProtocol, AbstractConnection): outbound_redis_connection = None # type: txredisapi.RedisProtocol def connectionMade(self): - logger.info("Connected to redis instance") - self.subscribe(self.stream_name) - self.send_command(ReplicateCommand()) - + logger.info("Connected to redis") + run_as_background_process("subscribe-replication", self._send_subscribe) self.handler.new_connection(self) + async def _send_subscribe(self): + # it's important to make sure that we only send the REPLICATE command once we + # have successfully subscribed to the stream - otherwise we might miss the + # POSITION response sent back by the other end. + logger.info("Sending redis SUBSCRIBE for %s", self.stream_name) + await make_deferred_yieldable(self.subscribe(self.stream_name)) + logger.info( + "Successfully subscribed to redis stream, sending REPLICATE command" + ) + await self._async_send_command(ReplicateCommand()) + logger.info("REPLICATE successfully sent") + def messageReceived(self, pattern: str, channel: str, message: str): """Received a message from redis. """ @@ -119,7 +135,7 @@ class RedisSubscriber(txredisapi.SubscriberProtocol, AbstractConnection): logger.warning("Unhandled command: %r", cmd) def connectionLost(self, reason): - logger.info("Lost connection to redis instance") + logger.info("Lost connection to redis") self.handler.lost_connection(self) def send_command(self, cmd: Command): @@ -128,6 +144,10 @@ class RedisSubscriber(txredisapi.SubscriberProtocol, AbstractConnection): Args: cmd (Command) """ + run_as_background_process("send-cmd", self._send_command, cmd) + + async def _async_send_command(self, cmd: Command): + """Encode a replication command and send it over our outbound connection""" string = "%s %s" % (cmd.NAME, cmd.to_line()) if "\n" in string: raise Exception("Unexpected newline in command: %r", string) @@ -138,15 +158,9 @@ class RedisSubscriber(txredisapi.SubscriberProtocol, AbstractConnection): # remote instances. tcp_outbound_commands_counter.labels(cmd.NAME, "redis").inc() - async def _send(): - with PreserveLoggingContext(): - # Note that we use the other connection as we can't send - # commands using the subscription connection. - await self.outbound_redis_connection.publish( - self.stream_name, encoded_string - ) - - run_as_background_process("send-cmd", _send) + await make_deferred_yieldable( + self.outbound_redis_connection.publish(self.stream_name, encoded_string) + ) class RedisDirectTcpReplicationClientFactory(txredisapi.SubscriberFactory): -- cgit 1.4.1 From 7f7eedbebba561771f39c22c78d1dfa1412283a5 Mon Sep 17 00:00:00 2001 From: Richard van der Hoff Date: Tue, 5 May 2020 19:32:35 +0100 Subject: Wait for a POSITION on the right connection before accepting RDATA ... otherwise we can believe we're up to date when we're not. --- synapse/replication/tcp/handler.py | 55 +++++++++++++++++++++++++------------- synapse/replication/tcp/redis.py | 2 +- 2 files changed, 38 insertions(+), 19 deletions(-) (limited to 'synapse/replication/tcp') diff --git a/synapse/replication/tcp/handler.py b/synapse/replication/tcp/handler.py index 0e8a38fd80..ffce3db629 100644 --- a/synapse/replication/tcp/handler.py +++ b/synapse/replication/tcp/handler.py @@ -81,9 +81,6 @@ class ReplicationCommandHandler: self._instance_id = hs.get_instance_id() self._instance_name = hs.get_instance_name() - # 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] @@ -103,6 +100,9 @@ class ReplicationCommandHandler: # outgoing replication commands to.) self._connections = [] # type: List[AbstractConnection] + # For each connection, the incoming streams that are coming from that connection + self._streams_by_connection = {} # type: Dict[AbstractConnection, Set[str]] + LaterGauge( "synapse_replication_tcp_resource_total_connections", "", @@ -258,12 +258,14 @@ class ReplicationCommandHandler: # 2. so we don't race with getting a POSITION command and fetching # missing RDATA. with await self._position_linearizer.queue(cmd.stream_name): - if stream_name not in self._streams_connected: - # If the stream isn't marked as connected then we haven't seen a - # `POSITION` command yet, and so we may have missed some rows. + # make sure that we've processed a POSITION for this stream *on this + # connection*. (A POSITION on another connection is no good, as there + # is no guarantee that we have seen all the intermediate updates.) + sbc = self._streams_by_connection.get(conn) + if not sbc or stream_name not in sbc: # Let's drop the row for now, on the assumption we'll receive a # `POSITION` soon and we'll catch up correctly then. - logger.warning( + logger.info( "Discarding RDATA for unconnected stream %s -> %s", stream_name, cmd.token, @@ -298,30 +300,33 @@ class ReplicationCommandHandler: # Ignore POSITION that are just our own echoes return - stream = self._streams.get(cmd.stream_name) + logger.info("Handling '%s %s'", cmd.NAME, cmd.to_line()) + + stream_name = cmd.stream_name + stream = self._streams.get(stream_name) if not stream: - logger.error("Got POSITION for unknown stream: %s", cmd.stream_name) + logger.error("Got POSITION for unknown stream: %s", 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): + with await self._position_linearizer.queue(stream_name): # We're about to go and catch up with the stream, so remove from set # of connected streams. - self._streams_connected.discard(cmd.stream_name) + for streams in self._streams_by_connection.values(): + streams.discard(stream_name) # We clear the pending batches for the stream as the fetching of the # missing updates below will fetch all rows in the batch. - self._pending_batches.pop(cmd.stream_name, []) + self._pending_batches.pop(stream_name, []) # Find where we previously streamed up to. current_token = self._replication_data_handler.get_streams_to_replicate().get( - cmd.stream_name + stream_name ) if current_token is None: logger.warning( - "Got POSITION for stream we're not subscribed to: %s", - cmd.stream_name, + "Got POSITION for stream we're not subscribed to: %s", stream_name, ) return @@ -330,6 +335,12 @@ class ReplicationCommandHandler: # between then and now. missing_updates = cmd.token != current_token while missing_updates: + logger.info( + "Fetching replication rows for '%s' between %i and %i", + stream_name, + current_token, + cmd.token, + ) ( updates, current_token, @@ -343,13 +354,15 @@ 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], + stream_name, token, [stream.parse_row(row) for row in rows], ) + logger.info("Caught up with stream '%s' to %i", stream_name, cmd.token) + # We've now caught up to position sent to us, notify handler. - await self._replication_data_handler.on_position(cmd.stream_name, cmd.token) + await self._replication_data_handler.on_position(stream_name, cmd.token) - self._streams_connected.add(cmd.stream_name) + self._streams_by_connection.setdefault(conn, set()).add(stream_name) async def on_REMOTE_SERVER_UP( self, conn: AbstractConnection, cmd: RemoteServerUpCommand @@ -407,6 +420,12 @@ class ReplicationCommandHandler: def lost_connection(self, connection: AbstractConnection): """Called when a connection is closed/lost. """ + # we no longer need _streams_by_connection for this connection. + streams = self._streams_by_connection.pop(connection, None) + if streams: + logger.info( + "Lost replication connection; streams now disconnected: %s", streams + ) try: self._connections.remove(connection) except ValueError: diff --git a/synapse/replication/tcp/redis.py b/synapse/replication/tcp/redis.py index 61dbf4ddbe..b3ae9f401c 100644 --- a/synapse/replication/tcp/redis.py +++ b/synapse/replication/tcp/redis.py @@ -144,7 +144,7 @@ class RedisSubscriber(txredisapi.SubscriberProtocol, AbstractConnection): Args: cmd (Command) """ - run_as_background_process("send-cmd", self._send_command, cmd) + run_as_background_process("send-cmd", self._async_send_command, cmd) async def _async_send_command(self, cmd: Command): """Encode a replication command and send it over our outbound connection""" -- cgit 1.4.1