diff options
Diffstat (limited to 'synapse/replication/tcp/protocol.py')
| -rw-r--r-- | synapse/replication/tcp/protocol.py | 135 |
1 files changed, 88 insertions, 47 deletions
diff --git a/synapse/replication/tcp/protocol.py b/synapse/replication/tcp/protocol.py index c870475cd1..74e892c104 100644 --- a/synapse/replication/tcp/protocol.py +++ b/synapse/replication/tcp/protocol.py @@ -49,29 +49,39 @@ indicate which side is sending, these are *not* included on the wire:: * connection closed by server * """ +import fcntl +import logging +import struct +from collections import defaultdict + +from six import iteritems, iterkeys + +from prometheus_client import Counter + from twisted.internet import defer from twisted.protocols.basic import LineOnlyReceiver from twisted.python.failure import Failure -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 synapse.metrics import LaterGauge +from synapse.metrics.background_process_metrics import run_as_background_process +from synapse.util.logcontext import make_deferred_yieldable, run_in_background from synapse.util.stringutils import random_string -from prometheus_client import Counter - -from collections import defaultdict - -from six import iterkeys, iteritems - -import logging -import struct -import fcntl +from .commands import ( + COMMAND_MAP, + VALID_CLIENT_COMMANDS, + VALID_SERVER_COMMANDS, + ErrorCommand, + NameCommand, + PingCommand, + PositionCommand, + RdataCommand, + ReplicateCommand, + ServerCommand, + SyncCommand, + UserSyncCommand, +) +from .streams import STREAMS_MAP connection_close_counter = Counter( "synapse_replication_tcp_protocol_close_reason", "", ["reason_type"]) @@ -214,7 +224,11 @@ class BaseReplicationStreamProtocol(LineOnlyReceiver): # Now lets try and call on_<CMD_NAME> function try: - getattr(self, "on_%s" % (cmd_name,))(cmd) + run_as_background_process( + "replication-" + cmd.get_logcontext_id(), + getattr(self, "on_%s" % (cmd_name,)), + cmd, + ) except Exception: logger.exception("[%s] Failed to handle line: %r", self.id(), line) @@ -379,7 +393,7 @@ class ServerReplicationStreamProtocol(BaseReplicationStreamProtocol): self.name = cmd.data def on_USER_SYNC(self, cmd): - self.streamer.on_user_sync( + return self.streamer.on_user_sync( self.conn_id, cmd.user_id, cmd.is_syncing, cmd.last_sync_ms, ) @@ -389,22 +403,33 @@ class ServerReplicationStreamProtocol(BaseReplicationStreamProtocol): if stream_name == "ALL": # Subscribe to all streams we're publishing to. - for stream in iterkeys(self.streamer.streams_by_name): - self.subscribe_to_stream(stream, token) + deferreds = [ + run_in_background( + self.subscribe_to_stream, + stream, token, + ) + for stream in iterkeys(self.streamer.streams_by_name) + ] + + return make_deferred_yieldable( + defer.gatherResults(deferreds, consumeErrors=True) + ) else: - self.subscribe_to_stream(stream_name, token) + return self.subscribe_to_stream(stream_name, token) def on_FEDERATION_ACK(self, cmd): - self.streamer.federation_ack(cmd.token) + return self.streamer.federation_ack(cmd.token) def on_REMOVE_PUSHER(self, cmd): - self.streamer.on_remove_pusher(cmd.app_id, cmd.push_key, cmd.user_id) + return self.streamer.on_remove_pusher( + cmd.app_id, cmd.push_key, cmd.user_id, + ) def on_INVALIDATE_CACHE(self, cmd): - self.streamer.on_invalidate_cache(cmd.cache_func, cmd.keys) + return self.streamer.on_invalidate_cache(cmd.cache_func, cmd.keys) def on_USER_IP(self, cmd): - self.streamer.on_user_ip( + return self.streamer.on_user_ip( cmd.user_id, cmd.access_token, cmd.ip, cmd.user_agent, cmd.device_id, cmd.last_seen, ) @@ -534,14 +559,13 @@ class ClientReplicationStreamProtocol(BaseReplicationStreamProtocol): # Check if this is the last of a batch of updates rows = self.pending_batches.pop(stream_name, []) rows.append(row) - - self.handler.on_rdata(stream_name, cmd.token, rows) + return self.handler.on_rdata(stream_name, cmd.token, rows) def on_POSITION(self, cmd): - self.handler.on_position(cmd.stream_name, cmd.token) + return self.handler.on_position(cmd.stream_name, cmd.token) def on_SYNC(self, cmd): - self.handler.on_sync(cmd.data) + return self.handler.on_sync(cmd.data) def replicate(self, stream_name, token): """Send the subscription request to the server @@ -564,11 +588,13 @@ class ClientReplicationStreamProtocol(BaseReplicationStreamProtocol): # The following simply registers metrics for the replication connections pending_commands = LaterGauge( - "pending_commands", "", ["name", "conn_id"], + "synapse_replication_tcp_protocol_pending_commands", + "", + ["name", "conn_id"], lambda: { - (p.name, p.conn_id): len(p.pending_commands) - for p in connected_connections - }) + (p.name, p.conn_id): len(p.pending_commands) for p in connected_connections + }, +) def transport_buffer_size(protocol): @@ -579,11 +605,13 @@ def transport_buffer_size(protocol): transport_send_buffer = LaterGauge( - "synapse_replication_tcp_transport_send_buffer", "", ["name", "conn_id"], + "synapse_replication_tcp_protocol_transport_send_buffer", + "", + ["name", "conn_id"], lambda: { - (p.name, p.conn_id): transport_buffer_size(p) - for p in connected_connections - }) + (p.name, p.conn_id): transport_buffer_size(p) for p in connected_connections + }, +) def transport_kernel_read_buffer_size(protocol, read=True): @@ -602,37 +630,50 @@ def transport_kernel_read_buffer_size(protocol, read=True): tcp_transport_kernel_send_buffer = LaterGauge( - "synapse_replication_tcp_transport_kernel_send_buffer", "", ["name", "conn_id"], + "synapse_replication_tcp_protocol_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 - }) + }, +) tcp_transport_kernel_read_buffer = LaterGauge( - "synapse_replication_tcp_transport_kernel_read_buffer", "", ["name", "conn_id"], + "synapse_replication_tcp_protocol_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 - }) + }, +) tcp_inbound_commands = LaterGauge( - "synapse_replication_tcp_inbound_commands", "", ["command", "name", "conn_id"], + "synapse_replication_tcp_protocol_inbound_commands", + "", + ["command", "name", "conn_id"], lambda: { (k[0], p.name, p.conn_id): count for p in connected_connections for k, count in iteritems(p.inbound_commands_counter) - }) + }, +) tcp_outbound_commands = LaterGauge( - "synapse_replication_tcp_outbound_commands", "", ["command", "name", "conn_id"], + "synapse_replication_tcp_protocol_outbound_commands", + "", + ["command", "name", "conn_id"], lambda: { (k[0], p.name, p.conn_id): count for p in connected_connections for k, count in iteritems(p.outbound_commands_counter) - }) + }, +) # number of updates received for each RDATA stream -inbound_rdata_count = Counter("synapse_replication_tcp_inbound_rdata_count", "", - ["stream_name"]) +inbound_rdata_count = Counter( + "synapse_replication_tcp_protocol_inbound_rdata_count", "", ["stream_name"] +) |