diff options
Diffstat (limited to 'synapse/replication/tcp/resource.py')
| -rw-r--r-- | synapse/replication/tcp/resource.py | 206 |
1 files changed, 35 insertions, 171 deletions
diff --git a/synapse/replication/tcp/resource.py b/synapse/replication/tcp/resource.py index d1e98428bc..41569305df 100644 --- a/synapse/replication/tcp/resource.py +++ b/synapse/replication/tcp/resource.py @@ -18,31 +18,17 @@ import logging import random -from six import itervalues - from prometheus_client import Counter -from twisted.internet import defer from twisted.internet.protocol import Factory -from synapse.metrics import LaterGauge from synapse.metrics.background_process_metrics import run_as_background_process -from synapse.util.metrics import Measure, measure_func - -from .protocol import ServerReplicationStreamProtocol -from .streams import STREAMS_MAP -from .streams.federation import FederationStream +from synapse.replication.tcp.protocol import ServerReplicationStreamProtocol +from synapse.util.metrics import Measure stream_updates_counter = Counter( "synapse_replication_tcp_resource_stream_updates", "", ["stream_name"] ) -user_sync_counter = Counter("synapse_replication_tcp_resource_user_sync", "") -federation_ack_counter = Counter("synapse_replication_tcp_resource_federation_ack", "") -remove_pusher_counter = Counter("synapse_replication_tcp_resource_remove_pusher", "") -invalidate_cache_counter = Counter( - "synapse_replication_tcp_resource_invalidate_cache", "" -) -user_ip_cache_counter = Counter("synapse_replication_tcp_resource_user_ip_cache", "") logger = logging.getLogger(__name__) @@ -52,13 +38,23 @@ class ReplicationStreamProtocolFactory(Factory): """ def __init__(self, hs): - self.streamer = ReplicationStreamer(hs) + self.command_handler = hs.get_tcp_replication() self.clock = hs.get_clock() self.server_name = hs.config.server_name + # If we've created a `ReplicationStreamProtocolFactory` then we're + # almost certainly registering a replication listener, so let's ensure + # that we've started a `ReplicationStreamer` instance to actually push + # data. + # + # (This is a bit of a weird place to do this, but the alternatives such + # as putting this in `HomeServer.setup()`, requires either passing the + # listener config again or always starting a `ReplicationStreamer`.) + hs.get_replication_streamer() + def buildProtocol(self, addr): return ServerReplicationStreamProtocol( - self.server_name, self.clock, self.streamer + self.server_name, self.clock, self.command_handler ) @@ -71,66 +67,22 @@ class ReplicationStreamer(object): def __init__(self, hs): self.store = hs.get_datastore() - self.presence_handler = hs.get_presence_handler() self.clock = hs.get_clock() self.notifier = hs.get_notifier() - self._server_notices_sender = hs.get_server_notices_sender() + self._instance_name = hs.get_instance_name() self._replication_torture_level = hs.config.replication_torture_level - # Current connections. - self.connections = [] - - LaterGauge( - "synapse_replication_tcp_resource_total_connections", - "", - [], - lambda: len(self.connections), - ) - - # List of streams that clients can subscribe to. - # We only support federation stream if federation sending hase been - # disabled on the master. - self.streams = [ - stream(hs) - for stream in itervalues(STREAMS_MAP) - if stream != FederationStream or not hs.config.send_federation - ] - - self.streams_by_name = {stream.NAME: stream for stream in self.streams} - - LaterGauge( - "synapse_replication_tcp_resource_connections_per_stream", - "", - ["stream_name"], - lambda: { - (stream_name,): len( - [ - conn - for conn in self.connections - if stream_name in conn.replication_streams - ] - ) - for stream_name in self.streams_by_name - }, - ) - - self.federation_sender = None - if not hs.config.send_federation: - self.federation_sender = hs.get_federation_sender() - self.notifier.add_replication_callback(self.on_notifier_poke) # Keeps track of whether we are currently checking for updates self.is_looping = False self.pending_updates = False - hs.get_reactor().addSystemEventTrigger("before", "shutdown", self.on_shutdown) + self.command_handler = hs.get_tcp_replication() - def on_shutdown(self): - # close all connections on shutdown - for conn in self.connections: - conn.send_error("server shutting down") + # Set of streams to replicate. + self.streams = self.command_handler.get_streams_to_replicate() def on_notifier_poke(self): """Checks if there is actually any new data and sends it to the @@ -139,7 +91,7 @@ class ReplicationStreamer(object): This should get called each time new data is available, even if it is currently being executed, so that nothing gets missed """ - if not self.connections: + if not self.command_handler.connected(): # Don't bother if nothing is listening. We still need to advance # the stream tokens otherwise they'll fall beihind forever for stream in self.streams: @@ -154,8 +106,7 @@ class ReplicationStreamer(object): run_as_background_process("replication_notifier", self._run_notifier_loop) - @defer.inlineCallbacks - def _run_notifier_loop(self): + async def _run_notifier_loop(self): self.is_looping = True try: @@ -166,11 +117,6 @@ class ReplicationStreamer(object): self.pending_updates = False with Measure(self.clock, "repl.stream.get_updates"): - # First we tell the streams that they should update their - # current tokens. - for stream in self.streams: - stream.advance_current_token() - all_streams = self.streams if self._replication_torture_level is not None: @@ -180,11 +126,13 @@ class ReplicationStreamer(object): random.shuffle(all_streams) for stream in all_streams: - if stream.last_token == stream.upto_token: + if stream.last_token == stream.current_token( + self._instance_name + ): continue if self._replication_torture_level: - yield self.clock.sleep( + await self.clock.sleep( self._replication_torture_level / 1000.0 ) @@ -192,18 +140,17 @@ class ReplicationStreamer(object): "Getting stream: %s: %s -> %s", stream.NAME, stream.last_token, - stream.upto_token, + stream.current_token(self._instance_name), ) try: - updates, current_token = yield stream.get_updates() + updates, current_token, limited = await stream.get_updates() + self.pending_updates |= limited except Exception: logger.info("Failed to handle stream %s", stream.NAME) raise logger.debug( - "Sending %d updates to %d connections", - len(updates), - len(self.connections), + "Sending %d updates", len(updates), ) if updates: @@ -219,102 +166,19 @@ class ReplicationStreamer(object): # token. See RdataCommand for more details. batched_updates = _batch_updates(updates) - for conn in self.connections: - for token, row in batched_updates: - try: - conn.stream_update(stream.NAME, token, row) - except Exception: - logger.exception("Failed to replicate") + for token, row in batched_updates: + try: + self.command_handler.stream_update( + stream.NAME, token, row + ) + except Exception: + logger.exception("Failed to replicate") logger.debug("No more pending updates, breaking poke loop") finally: self.pending_updates = False self.is_looping = False - @measure_func("repl.get_stream_updates") - def get_stream_updates(self, stream_name, token): - """For a given stream get all updates since token. This is called when - a client first subscribes to a stream. - """ - stream = self.streams_by_name.get(stream_name, None) - if not stream: - raise Exception("unknown stream %s", stream_name) - - return stream.get_updates_since(token) - - @measure_func("repl.federation_ack") - def federation_ack(self, token): - """We've received an ack for federation stream from a client. - """ - federation_ack_counter.inc() - if self.federation_sender: - self.federation_sender.federation_ack(token) - - @measure_func("repl.on_user_sync") - @defer.inlineCallbacks - def on_user_sync(self, conn_id, user_id, is_syncing, last_sync_ms): - """A client has started/stopped syncing on a worker. - """ - user_sync_counter.inc() - yield self.presence_handler.update_external_syncs_row( - conn_id, user_id, is_syncing, last_sync_ms - ) - - @measure_func("repl.on_remove_pusher") - @defer.inlineCallbacks - def on_remove_pusher(self, app_id, push_key, user_id): - """A client has asked us to remove a pusher - """ - remove_pusher_counter.inc() - yield self.store.delete_pusher_by_app_id_pushkey_user_id( - app_id=app_id, pushkey=push_key, user_id=user_id - ) - - self.notifier.on_new_replication_data() - - @measure_func("repl.on_invalidate_cache") - def on_invalidate_cache(self, cache_func, keys): - """The client has asked us to invalidate a cache - """ - invalidate_cache_counter.inc() - getattr(self.store, cache_func).invalidate(tuple(keys)) - - @measure_func("repl.on_user_ip") - @defer.inlineCallbacks - def on_user_ip(self, user_id, access_token, ip, user_agent, device_id, last_seen): - """The client saw a user request - """ - user_ip_cache_counter.inc() - yield self.store.insert_client_ip( - user_id, access_token, ip, user_agent, device_id, last_seen - ) - yield self._server_notices_sender.on_user_ip(user_id) - - def send_sync_to_all_connections(self, data): - """Sends a SYNC command to all clients. - - Used in tests. - """ - for conn in self.connections: - conn.send_sync(data) - - def new_connection(self, connection): - """A new client connection has been established - """ - self.connections.append(connection) - - def lost_connection(self, connection): - """A client connection has been lost - """ - try: - self.connections.remove(connection) - except ValueError: - pass - - # We need to tell the presence handler that the connection has been - # lost so that it can handle any ongoing syncs on that connection. - self.presence_handler.update_external_syncs_clear(connection.conn_id) - def _batch_updates(updates): """Takes a list of updates of form [(token, row)] and sets the token to |