diff options
Diffstat (limited to 'synapse/replication')
| -rw-r--r-- | synapse/replication/http/send_event.py | 20 | ||||
| -rw-r--r-- | synapse/replication/tcp/commands.py | 18 | ||||
| -rw-r--r-- | synapse/replication/tcp/protocol.py | 105 | ||||
| -rw-r--r-- | synapse/replication/tcp/resource.py | 43 |
4 files changed, 85 insertions, 101 deletions
diff --git a/synapse/replication/http/send_event.py b/synapse/replication/http/send_event.py index bbe2f967b7..a9baa2c1c3 100644 --- a/synapse/replication/http/send_event.py +++ b/synapse/replication/http/send_event.py @@ -23,7 +23,6 @@ from synapse.events.snapshot import EventContext from synapse.http.servlet import RestServlet, parse_json_object_from_request from synapse.util.async import sleep from synapse.util.caches.response_cache import ResponseCache -from synapse.util.logcontext import make_deferred_yieldable, preserve_fn from synapse.util.metrics import Measure from synapse.types import Requester, UserID @@ -115,20 +114,15 @@ class ReplicationSendEventRestServlet(RestServlet): self.clock = hs.get_clock() # The responses are tiny, so we may as well cache them for a while - self.response_cache = ResponseCache(hs, timeout_ms=30 * 60 * 1000) + self.response_cache = ResponseCache(hs, "send_event", timeout_ms=30 * 60 * 1000) def on_PUT(self, request, event_id): - result = self.response_cache.get(event_id) - if not result: - result = self.response_cache.set( - event_id, - self._handle_request(request) - ) - else: - logger.warn("Returning cached response") - return make_deferred_yieldable(result) - - @preserve_fn + return self.response_cache.wrap( + event_id, + self._handle_request, + request + ) + @defer.inlineCallbacks def _handle_request(self, request): with Measure(self.clock, "repl_send_event_parse"): diff --git a/synapse/replication/tcp/commands.py b/synapse/replication/tcp/commands.py index 171227cce2..12aac3cc6b 100644 --- a/synapse/replication/tcp/commands.py +++ b/synapse/replication/tcp/commands.py @@ -19,11 +19,13 @@ allowed to be sent by which side. """ import logging -import ujson as json +import simplejson logger = logging.getLogger(__name__) +_json_encoder = simplejson.JSONEncoder(namedtuple_as_object=False) + class Command(object): """The base command class. @@ -100,14 +102,14 @@ class RdataCommand(Command): return cls( stream_name, None if token == "batch" else int(token), - json.loads(row_json) + simplejson.loads(row_json) ) def to_line(self): return " ".join(( self.stream_name, str(self.token) if self.token is not None else "batch", - json.dumps(self.row), + _json_encoder.encode(self.row), )) @@ -298,10 +300,12 @@ class InvalidateCacheCommand(Command): def from_line(cls, line): cache_func, keys_json = line.split(" ", 1) - return cls(cache_func, json.loads(keys_json)) + return cls(cache_func, simplejson.loads(keys_json)) def to_line(self): - return " ".join((self.cache_func, json.dumps(self.keys))) + return " ".join(( + self.cache_func, _json_encoder.encode(self.keys), + )) class UserIpCommand(Command): @@ -325,14 +329,14 @@ class UserIpCommand(Command): def from_line(cls, line): user_id, jsn = line.split(" ", 1) - access_token, ip, user_agent, device_id, last_seen = json.loads(jsn) + access_token, ip, user_agent, device_id, last_seen = simplejson.loads(jsn) return cls( user_id, access_token, ip, user_agent, device_id, last_seen ) def to_line(self): - return self.user_id + " " + json.dumps(( + return self.user_id + " " + _json_encoder.encode(( self.access_token, self.ip, self.user_agent, self.device_id, self.last_seen, )) diff --git a/synapse/replication/tcp/protocol.py b/synapse/replication/tcp/protocol.py index 0a9a290af4..a6280aae70 100644 --- a/synapse/replication/tcp/protocol.py +++ b/synapse/replication/tcp/protocol.py @@ -53,28 +53,28 @@ from twisted.internet import defer from twisted.protocols.basic import LineOnlyReceiver from twisted.python.failure import Failure -from commands import ( +from .commands import ( COMMAND_MAP, VALID_CLIENT_COMMANDS, VALID_SERVER_COMMANDS, ErrorCommand, ServerCommand, RdataCommand, PositionCommand, PingCommand, NameCommand, ReplicateCommand, UserSyncCommand, SyncCommand, ) -from streams import STREAMS_MAP +from .streams import STREAMS_MAP +from synapse.metrics import LaterGauge from synapse.util.stringutils import random_string -from synapse.metrics.metric import CounterMetric -import logging -import synapse.metrics -import struct -import fcntl +from prometheus_client import Counter +from collections import defaultdict -metrics = synapse.metrics.get_metrics_for(__name__) +from six import iterkeys, iteritems -connection_close_counter = metrics.register_counter( - "close_reason", labels=["reason_type"], -) +import logging +import struct +import fcntl +connection_close_counter = Counter( + "synapse_replication_tcp_protocol_close_reason", "", ["reason_type"]) # A list of all connected protocols. This allows us to send metrics about the # connections. @@ -136,12 +136,8 @@ class BaseReplicationStreamProtocol(LineOnlyReceiver): # The LoopingCall for sending pings. self._send_ping_loop = None - self.inbound_commands_counter = CounterMetric( - "inbound_commands", labels=["command"], - ) - self.outbound_commands_counter = CounterMetric( - "outbound_commands", labels=["command"], - ) + self.inbound_commands_counter = defaultdict(int) + self.outbound_commands_counter = defaultdict(int) def connectionMade(self): logger.info("[%s] Connection established", self.id()) @@ -201,7 +197,8 @@ class BaseReplicationStreamProtocol(LineOnlyReceiver): self.last_received_command = self.clock.time_msec() - self.inbound_commands_counter.inc(cmd_name) + self.inbound_commands_counter[cmd_name] = ( + self.inbound_commands_counter[cmd_name] + 1) cmd_cls = COMMAND_MAP[cmd_name] try: @@ -251,8 +248,8 @@ class BaseReplicationStreamProtocol(LineOnlyReceiver): self._queue_command(cmd) return - self.outbound_commands_counter.inc(cmd.NAME) - + self.outbound_commands_counter[cmd.NAME] = ( + self.outbound_commands_counter[cmd.NAME] + 1) string = "%s %s" % (cmd.NAME, cmd.to_line(),) if "\n" in string: raise Exception("Unexpected newline in command: %r", string) @@ -317,9 +314,9 @@ class BaseReplicationStreamProtocol(LineOnlyReceiver): def connectionLost(self, reason): logger.info("[%s] Replication connection closed: %r", self.id(), reason) if isinstance(reason, Failure): - connection_close_counter.inc(reason.type.__name__) + connection_close_counter.labels(reason.type.__name__).inc() else: - connection_close_counter.inc(reason.__class__.__name__) + connection_close_counter.labels(reason.__class__.__name__).inc() try: # Remove us from list of connections to be monitored @@ -392,7 +389,7 @@ class ServerReplicationStreamProtocol(BaseReplicationStreamProtocol): if stream_name == "ALL": # Subscribe to all streams we're publishing to. - for stream in self.streamer.streams_by_name.iterkeys(): + for stream in iterkeys(self.streamer.streams_by_name): self.subscribe_to_stream(stream, token) else: self.subscribe_to_stream(stream_name, token) @@ -498,7 +495,7 @@ class ClientReplicationStreamProtocol(BaseReplicationStreamProtocol): BaseReplicationStreamProtocol.connectionMade(self) # Once we've connected subscribe to the necessary streams - for stream_name, token in self.handler.get_streams_to_replicate().iteritems(): + for stream_name, token in iteritems(self.handler.get_streams_to_replicate()): self.replicate(stream_name, token) # Tell the server if we have any users currently syncing (should only @@ -518,7 +515,7 @@ class ClientReplicationStreamProtocol(BaseReplicationStreamProtocol): def on_RDATA(self, cmd): stream_name = cmd.stream_name - inbound_rdata_count.inc(stream_name) + inbound_rdata_count.labels(stream_name).inc() try: row = STREAMS_MAP[stream_name].ROW_TYPE(*cmd.row) @@ -566,14 +563,12 @@ class ClientReplicationStreamProtocol(BaseReplicationStreamProtocol): # The following simply registers metrics for the replication connections -metrics.register_callback( - "pending_commands", +pending_commands = LaterGauge( + "pending_commands", "", ["name", "conn_id"], lambda: { (p.name, p.conn_id): len(p.pending_commands) for p in connected_connections - }, - labels=["name", "conn_id"], -) + }) def transport_buffer_size(protocol): @@ -583,14 +578,12 @@ def transport_buffer_size(protocol): return 0 -metrics.register_callback( - "transport_send_buffer", +transport_send_buffer = LaterGauge( + "synapse_replication_tcp_transport_send_buffer", "", ["name", "conn_id"], lambda: { (p.name, p.conn_id): transport_buffer_size(p) for p in connected_connections - }, - labels=["name", "conn_id"], -) + }) def transport_kernel_read_buffer_size(protocol, read=True): @@ -608,48 +601,38 @@ def transport_kernel_read_buffer_size(protocol, read=True): return 0 -metrics.register_callback( - "transport_kernel_send_buffer", +tcp_transport_kernel_send_buffer = LaterGauge( + "synapse_replication_tcp_transport_kernel_send_buffer", "", ["name", "conn_id"], lambda: { (p.name, p.conn_id): transport_kernel_read_buffer_size(p, False) for p in connected_connections - }, - labels=["name", "conn_id"], -) + }) -metrics.register_callback( - "transport_kernel_read_buffer", +tcp_transport_kernel_read_buffer = LaterGauge( + "synapse_replication_tcp_transport_kernel_read_buffer", "", ["name", "conn_id"], lambda: { (p.name, p.conn_id): transport_kernel_read_buffer_size(p, True) for p in connected_connections - }, - labels=["name", "conn_id"], -) + }) -metrics.register_callback( - "inbound_commands", +tcp_inbound_commands = LaterGauge( + "synapse_replication_tcp_inbound_commands", "", ["command", "name", "conn_id"], lambda: { (k[0], p.name, p.conn_id): count for p in connected_connections - for k, count in p.inbound_commands_counter.counts.iteritems() - }, - labels=["command", "name", "conn_id"], -) + for k, count in iteritems(p.inbound_commands_counter.counts) + }) -metrics.register_callback( - "outbound_commands", +tcp_outbound_commands = LaterGauge( + "synapse_replication_tcp_outbound_commands", "", ["command", "name", "conn_id"], lambda: { (k[0], p.name, p.conn_id): count for p in connected_connections - for k, count in p.outbound_commands_counter.counts.iteritems() - }, - labels=["command", "name", "conn_id"], -) + for k, count in iteritems(p.outbound_commands_counter.counts) + }) # number of updates received for each RDATA stream -inbound_rdata_count = metrics.register_counter( - "inbound_rdata_count", - labels=["stream_name"], -) +inbound_rdata_count = Counter("synapse_replication_tcp_inbound_rdata_count", "", + ["stream_name"]) diff --git a/synapse/replication/tcp/resource.py b/synapse/replication/tcp/resource.py index 786c3fe864..63bd6d2652 100644 --- a/synapse/replication/tcp/resource.py +++ b/synapse/replication/tcp/resource.py @@ -18,24 +18,25 @@ from twisted.internet import defer, reactor from twisted.internet.protocol import Factory -from streams import STREAMS_MAP, FederationStream -from protocol import ServerReplicationStreamProtocol +from .streams import STREAMS_MAP, FederationStream +from .protocol import ServerReplicationStreamProtocol from synapse.util.metrics import Measure, measure_func +from synapse.metrics import LaterGauge import logging -import synapse.metrics +from prometheus_client import Counter +from six import itervalues -metrics = synapse.metrics.get_metrics_for(__name__) -stream_updates_counter = metrics.register_counter( - "stream_updates", labels=["stream_name"] -) -user_sync_counter = metrics.register_counter("user_sync") -federation_ack_counter = metrics.register_counter("federation_ack") -remove_pusher_counter = metrics.register_counter("remove_pusher") -invalidate_cache_counter = metrics.register_counter("invalidate_cache") -user_ip_cache_counter = metrics.register_counter("user_ip_cache") +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__) @@ -69,33 +70,34 @@ class ReplicationStreamer(object): 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() # Current connections. self.connections = [] - metrics.register_callback("total_connections", lambda: len(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 STREAMS_MAP.itervalues() + 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} - metrics.register_callback( - "connections_per_stream", + 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 - }, - labels=["stream_name"], - ) + }) self.federation_sender = None if not hs.config.send_federation: @@ -175,7 +177,7 @@ class ReplicationStreamer(object): logger.info( "Streaming: %s -> %s", stream.NAME, updates[-1][0] ) - stream_updates_counter.inc_by(len(updates), stream.NAME) + stream_updates_counter.labels(stream.NAME).inc(len(updates)) # Some streams return multiple rows with the same stream IDs, # we need to make sure they get sent out in batches. We do @@ -253,6 +255,7 @@ class ReplicationStreamer(object): 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. |