2019-03-13 20:02:56 +00:00
|
|
|
# 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.
|
|
|
|
|
2021-03-29 15:43:20 +00:00
|
|
|
import abc
|
2019-03-13 20:02:56 +00:00
|
|
|
import logging
|
2021-07-15 10:52:56 +00:00
|
|
|
from collections import OrderedDict
|
2021-04-28 10:04:38 +00:00
|
|
|
from typing import TYPE_CHECKING, Dict, Hashable, Iterable, List, Optional, Set, Tuple
|
2019-03-13 20:02:56 +00:00
|
|
|
|
2021-07-15 10:52:56 +00:00
|
|
|
import attr
|
2019-03-13 20:02:56 +00:00
|
|
|
from prometheus_client import Counter
|
2021-07-15 10:52:56 +00:00
|
|
|
from typing_extensions import Literal
|
2019-03-13 20:02:56 +00:00
|
|
|
|
2021-04-28 10:04:38 +00:00
|
|
|
from twisted.internet import defer
|
2021-09-10 16:03:18 +00:00
|
|
|
from twisted.internet.interfaces import IDelayedCall
|
2021-04-28 10:04:38 +00:00
|
|
|
|
2019-03-13 20:02:56 +00:00
|
|
|
import synapse.metrics
|
2020-08-17 11:24:46 +00:00
|
|
|
from synapse.api.presence import UserPresenceState
|
2020-02-07 13:56:38 +00:00
|
|
|
from synapse.events import EventBase
|
2019-03-13 20:02:56 +00:00
|
|
|
from synapse.federation.sender.per_destination_queue import PerDestinationQueue
|
|
|
|
from synapse.federation.sender.transaction_manager import TransactionManager
|
|
|
|
from synapse.federation.units import Edu
|
2021-04-28 10:04:38 +00:00
|
|
|
from synapse.logging.context import make_deferred_yieldable, run_in_background
|
2019-03-13 20:02:56 +00:00
|
|
|
from synapse.metrics import (
|
|
|
|
LaterGauge,
|
|
|
|
event_processing_loop_counter,
|
|
|
|
event_processing_loop_room_count,
|
|
|
|
events_processed_counter,
|
|
|
|
)
|
2021-07-15 10:52:56 +00:00
|
|
|
from synapse.metrics.background_process_metrics import (
|
|
|
|
run_as_background_process,
|
|
|
|
wrap_as_background_process,
|
|
|
|
)
|
2021-04-22 15:43:50 +00:00
|
|
|
from synapse.types import JsonDict, ReadReceipt, RoomStreamToken
|
2021-07-15 10:52:56 +00:00
|
|
|
from synapse.util import Clock
|
2021-04-19 09:50:49 +00:00
|
|
|
from synapse.util.metrics import Measure
|
2019-03-13 20:02:56 +00:00
|
|
|
|
2021-03-29 15:43:20 +00:00
|
|
|
if TYPE_CHECKING:
|
2021-04-06 13:38:30 +00:00
|
|
|
from synapse.events.presence_router import PresenceRouter
|
2021-03-29 15:43:20 +00:00
|
|
|
from synapse.server import HomeServer
|
|
|
|
|
2019-03-13 20:02:56 +00:00
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
|
|
|
|
sent_pdus_destination_dist_count = Counter(
|
|
|
|
"synapse_federation_client_sent_pdu_destinations:count",
|
|
|
|
"Number of PDUs queued for sending to one or more destinations",
|
|
|
|
)
|
|
|
|
|
|
|
|
sent_pdus_destination_dist_total = Counter(
|
|
|
|
"synapse_federation_client_sent_pdu_destinations:total",
|
2019-11-21 12:00:14 +00:00
|
|
|
"Total number of PDUs queued for sending across all destinations",
|
2019-03-13 20:02:56 +00:00
|
|
|
)
|
|
|
|
|
2020-09-18 13:59:13 +00:00
|
|
|
# Time (in s) after Synapse's startup that we will begin to wake up destinations
|
|
|
|
# that have catch-up outstanding.
|
|
|
|
CATCH_UP_STARTUP_DELAY_SEC = 15
|
|
|
|
|
|
|
|
# Time (in s) to wait in between waking up each destination, i.e. one destination
|
|
|
|
# will be woken up every <x> seconds after Synapse's startup until we have woken
|
|
|
|
# every destination has outstanding catch-up.
|
|
|
|
CATCH_UP_STARTUP_INTERVAL_SEC = 5
|
|
|
|
|
2019-03-13 20:02:56 +00:00
|
|
|
|
2021-03-29 15:43:20 +00:00
|
|
|
class AbstractFederationSender(metaclass=abc.ABCMeta):
|
|
|
|
@abc.abstractmethod
|
|
|
|
def notify_new_events(self, max_token: RoomStreamToken) -> None:
|
|
|
|
"""This gets called when we have some new events we might want to
|
|
|
|
send out to other servers.
|
|
|
|
"""
|
|
|
|
raise NotImplementedError()
|
|
|
|
|
|
|
|
@abc.abstractmethod
|
|
|
|
async def send_read_receipt(self, receipt: ReadReceipt) -> None:
|
|
|
|
"""Send a RR to any other servers in the room
|
|
|
|
|
|
|
|
Args:
|
|
|
|
receipt: receipt to be sent
|
|
|
|
"""
|
|
|
|
raise NotImplementedError()
|
|
|
|
|
|
|
|
@abc.abstractmethod
|
|
|
|
def send_presence_to_destinations(
|
|
|
|
self, states: Iterable[UserPresenceState], destinations: Iterable[str]
|
|
|
|
) -> None:
|
|
|
|
"""Send the given presence states to the given destinations.
|
|
|
|
|
|
|
|
Args:
|
|
|
|
destinations:
|
|
|
|
"""
|
|
|
|
raise NotImplementedError()
|
|
|
|
|
|
|
|
@abc.abstractmethod
|
|
|
|
def build_and_send_edu(
|
|
|
|
self,
|
|
|
|
destination: str,
|
|
|
|
edu_type: str,
|
|
|
|
content: JsonDict,
|
|
|
|
key: Optional[Hashable] = None,
|
|
|
|
) -> None:
|
|
|
|
"""Construct an Edu object, and queue it for sending
|
|
|
|
|
|
|
|
Args:
|
|
|
|
destination: name of server to send to
|
|
|
|
edu_type: type of EDU to send
|
|
|
|
content: content of EDU
|
|
|
|
key: clobbering key for this edu
|
|
|
|
"""
|
|
|
|
raise NotImplementedError()
|
|
|
|
|
|
|
|
@abc.abstractmethod
|
|
|
|
def send_device_messages(self, destination: str) -> None:
|
|
|
|
raise NotImplementedError()
|
|
|
|
|
|
|
|
@abc.abstractmethod
|
|
|
|
def wake_destination(self, destination: str) -> None:
|
|
|
|
"""Called when we want to retry sending transactions to a remote.
|
|
|
|
|
|
|
|
This is mainly useful if the remote server has been down and we think it
|
|
|
|
might have come back.
|
|
|
|
"""
|
|
|
|
raise NotImplementedError()
|
|
|
|
|
|
|
|
@abc.abstractmethod
|
|
|
|
def get_current_token(self) -> int:
|
|
|
|
raise NotImplementedError()
|
|
|
|
|
|
|
|
@abc.abstractmethod
|
|
|
|
def federation_ack(self, instance_name: str, token: int) -> None:
|
|
|
|
raise NotImplementedError()
|
|
|
|
|
|
|
|
@abc.abstractmethod
|
|
|
|
async def get_replication_rows(
|
|
|
|
self, instance_name: str, from_token: int, to_token: int, target_row_count: int
|
|
|
|
) -> Tuple[List[Tuple[int, Tuple]], int, bool]:
|
|
|
|
raise NotImplementedError()
|
|
|
|
|
|
|
|
|
2021-07-15 10:52:56 +00:00
|
|
|
@attr.s
|
|
|
|
class _PresenceQueue:
|
|
|
|
"""A queue of destinations that need to be woken up due to new presence
|
|
|
|
updates.
|
|
|
|
|
|
|
|
Staggers waking up of per destination queues to ensure that we don't attempt
|
|
|
|
to start TLS connections with many hosts all at once, leading to pinned CPU.
|
|
|
|
"""
|
|
|
|
|
|
|
|
# The maximum duration in seconds between queuing up a destination and it
|
|
|
|
# being woken up.
|
|
|
|
_MAX_TIME_IN_QUEUE = 30.0
|
|
|
|
|
|
|
|
# The maximum duration in seconds between waking up consecutive destination
|
|
|
|
# queues.
|
|
|
|
_MAX_DELAY = 0.1
|
|
|
|
|
|
|
|
sender: "FederationSender" = attr.ib()
|
|
|
|
clock: Clock = attr.ib()
|
|
|
|
queue: "OrderedDict[str, Literal[None]]" = attr.ib(factory=OrderedDict)
|
|
|
|
processing: bool = attr.ib(default=False)
|
|
|
|
|
|
|
|
def add_to_queue(self, destination: str) -> None:
|
|
|
|
"""Add a destination to the queue to be woken up."""
|
|
|
|
|
|
|
|
self.queue[destination] = None
|
|
|
|
|
|
|
|
if not self.processing:
|
|
|
|
self._handle()
|
|
|
|
|
|
|
|
@wrap_as_background_process("_PresenceQueue.handle")
|
|
|
|
async def _handle(self) -> None:
|
|
|
|
"""Background process to drain the queue."""
|
|
|
|
|
|
|
|
if not self.queue:
|
|
|
|
return
|
|
|
|
|
|
|
|
assert not self.processing
|
|
|
|
self.processing = True
|
|
|
|
|
|
|
|
try:
|
|
|
|
# We start with a delay that should drain the queue quickly enough that
|
|
|
|
# we process all destinations in the queue in _MAX_TIME_IN_QUEUE
|
|
|
|
# seconds.
|
|
|
|
#
|
|
|
|
# We also add an upper bound to the delay, to gracefully handle the
|
|
|
|
# case where the queue only has a few entries in it.
|
|
|
|
current_sleep_seconds = min(
|
|
|
|
self._MAX_DELAY, self._MAX_TIME_IN_QUEUE / len(self.queue)
|
|
|
|
)
|
|
|
|
|
|
|
|
while self.queue:
|
|
|
|
destination, _ = self.queue.popitem(last=False)
|
|
|
|
|
|
|
|
queue = self.sender._get_per_destination_queue(destination)
|
|
|
|
|
|
|
|
if not queue._new_data_to_send:
|
|
|
|
# The per destination queue has already been woken up.
|
|
|
|
continue
|
|
|
|
|
|
|
|
queue.attempt_new_transaction()
|
|
|
|
|
|
|
|
await self.clock.sleep(current_sleep_seconds)
|
|
|
|
|
|
|
|
if not self.queue:
|
|
|
|
break
|
|
|
|
|
|
|
|
# More destinations may have been added to the queue, so we may
|
|
|
|
# need to reduce the delay to ensure everything gets processed
|
|
|
|
# within _MAX_TIME_IN_QUEUE seconds.
|
|
|
|
current_sleep_seconds = min(
|
|
|
|
current_sleep_seconds, self._MAX_TIME_IN_QUEUE / len(self.queue)
|
|
|
|
)
|
|
|
|
|
|
|
|
finally:
|
|
|
|
self.processing = False
|
|
|
|
|
|
|
|
|
2021-03-29 15:43:20 +00:00
|
|
|
class FederationSender(AbstractFederationSender):
|
|
|
|
def __init__(self, hs: "HomeServer"):
|
2019-03-13 20:02:56 +00:00
|
|
|
self.hs = hs
|
|
|
|
self.server_name = hs.hostname
|
|
|
|
|
|
|
|
self.store = hs.get_datastore()
|
|
|
|
self.state = hs.get_state_handler()
|
|
|
|
|
|
|
|
self.clock = hs.get_clock()
|
|
|
|
self.is_mine_id = hs.is_mine_id
|
|
|
|
|
2021-07-15 10:02:43 +00:00
|
|
|
self._presence_router: Optional["PresenceRouter"] = None
|
2019-03-13 20:02:56 +00:00
|
|
|
self._transaction_manager = TransactionManager(hs)
|
|
|
|
|
2020-07-10 17:26:36 +00:00
|
|
|
self._instance_name = hs.get_instance_name()
|
2020-07-29 22:22:13 +00:00
|
|
|
self._federation_shard_config = hs.config.worker.federation_shard_config
|
2020-07-10 17:26:36 +00:00
|
|
|
|
2019-03-13 20:02:56 +00:00
|
|
|
# map from destination to PerDestinationQueue
|
2021-07-15 10:02:43 +00:00
|
|
|
self._per_destination_queues: Dict[str, PerDestinationQueue] = {}
|
2019-03-13 20:02:56 +00:00
|
|
|
|
|
|
|
LaterGauge(
|
|
|
|
"synapse_federation_transaction_queue_pending_destinations",
|
|
|
|
"",
|
|
|
|
[],
|
|
|
|
lambda: sum(
|
|
|
|
1
|
|
|
|
for d in self._per_destination_queues.values()
|
|
|
|
if d.transmission_loop_running
|
|
|
|
),
|
|
|
|
)
|
|
|
|
|
|
|
|
LaterGauge(
|
|
|
|
"synapse_federation_transaction_queue_pending_pdus",
|
|
|
|
"",
|
|
|
|
[],
|
|
|
|
lambda: sum(
|
|
|
|
d.pending_pdu_count() for d in self._per_destination_queues.values()
|
|
|
|
),
|
|
|
|
)
|
|
|
|
LaterGauge(
|
|
|
|
"synapse_federation_transaction_queue_pending_edus",
|
|
|
|
"",
|
|
|
|
[],
|
|
|
|
lambda: sum(
|
|
|
|
d.pending_edu_count() for d in self._per_destination_queues.values()
|
|
|
|
),
|
|
|
|
)
|
|
|
|
|
|
|
|
self._is_processing = False
|
|
|
|
self._last_poked_id = -1
|
|
|
|
|
2019-03-20 16:02:25 +00:00
|
|
|
# map from room_id to a set of PerDestinationQueues which we believe are
|
|
|
|
# awaiting a call to flush_read_receipts_for_room. The presence of an entry
|
|
|
|
# here for a given room means that we are rate-limiting RR flushes to that room,
|
|
|
|
# and that there is a pending call to _flush_rrs_for_room in the system.
|
2021-07-15 10:02:43 +00:00
|
|
|
self._queues_awaiting_rr_flush_by_room: Dict[str, Set[PerDestinationQueue]] = {}
|
2019-03-20 16:02:25 +00:00
|
|
|
|
|
|
|
self._rr_txn_interval_per_room_ms = (
|
2021-09-13 17:07:12 +00:00
|
|
|
1000.0
|
|
|
|
/ hs.config.ratelimiting.federation_rr_transactions_per_room_per_second
|
2019-03-20 16:02:25 +00:00
|
|
|
)
|
|
|
|
|
2020-09-18 13:59:13 +00:00
|
|
|
# wake up destinations that have outstanding PDUs to be caught up
|
2021-09-10 16:03:18 +00:00
|
|
|
self._catchup_after_startup_timer: Optional[
|
|
|
|
IDelayedCall
|
|
|
|
] = self.clock.call_later(
|
2020-09-18 13:59:13 +00:00
|
|
|
CATCH_UP_STARTUP_DELAY_SEC,
|
|
|
|
run_as_background_process,
|
|
|
|
"wake_destinations_needing_catchup",
|
|
|
|
self._wake_destinations_needing_catchup,
|
|
|
|
)
|
|
|
|
|
2021-01-26 13:57:31 +00:00
|
|
|
self._external_cache = hs.get_external_cache()
|
|
|
|
|
2021-07-15 10:52:56 +00:00
|
|
|
self._presence_queue = _PresenceQueue(self, self.clock)
|
|
|
|
|
2020-02-07 13:56:38 +00:00
|
|
|
def _get_per_destination_queue(self, destination: str) -> PerDestinationQueue:
|
2019-03-20 16:02:25 +00:00
|
|
|
"""Get or create a PerDestinationQueue for the given destination
|
|
|
|
|
|
|
|
Args:
|
2020-02-07 13:56:38 +00:00
|
|
|
destination: server_name of remote server
|
2019-03-20 16:02:25 +00:00
|
|
|
"""
|
2019-03-13 20:02:56 +00:00
|
|
|
queue = self._per_destination_queues.get(destination)
|
|
|
|
if not queue:
|
|
|
|
queue = PerDestinationQueue(self.hs, self._transaction_manager, destination)
|
|
|
|
self._per_destination_queues[destination] = queue
|
|
|
|
return queue
|
|
|
|
|
2020-10-14 12:27:51 +00:00
|
|
|
def notify_new_events(self, max_token: RoomStreamToken) -> None:
|
2019-03-13 20:02:56 +00:00
|
|
|
"""This gets called when we have some new events we might want to
|
|
|
|
send out to other servers.
|
|
|
|
"""
|
2020-10-14 12:27:51 +00:00
|
|
|
# We just use the minimum stream ordering and ignore the vector clock
|
|
|
|
# component. This is safe to do as long as we *always* ignore the vector
|
|
|
|
# clock components.
|
|
|
|
current_id = max_token.stream
|
|
|
|
|
2019-03-13 20:02:56 +00:00
|
|
|
self._last_poked_id = max(current_id, self._last_poked_id)
|
|
|
|
|
|
|
|
if self._is_processing:
|
|
|
|
return
|
|
|
|
|
|
|
|
# fire off a processing loop in the background
|
|
|
|
run_as_background_process(
|
|
|
|
"process_event_queue_for_federation", self._process_event_queue_loop
|
|
|
|
)
|
|
|
|
|
2020-02-07 13:56:38 +00:00
|
|
|
async def _process_event_queue_loop(self) -> None:
|
2019-03-13 20:02:56 +00:00
|
|
|
try:
|
|
|
|
self._is_processing = True
|
|
|
|
while True:
|
2020-02-07 13:56:38 +00:00
|
|
|
last_token = await self.store.get_federation_out_pos("events")
|
|
|
|
next_token, events = await self.store.get_all_new_events_stream(
|
2019-03-13 20:02:56 +00:00
|
|
|
last_token, self._last_poked_id, limit=100
|
|
|
|
)
|
|
|
|
|
|
|
|
logger.debug("Handling %s -> %s", last_token, next_token)
|
|
|
|
|
|
|
|
if not events and next_token >= self._last_poked_id:
|
|
|
|
break
|
|
|
|
|
2021-04-28 10:04:38 +00:00
|
|
|
async def handle_event(event: EventBase) -> None:
|
2019-03-13 20:02:56 +00:00
|
|
|
# Only send events for this server.
|
|
|
|
send_on_behalf_of = event.internal_metadata.get_send_on_behalf_of()
|
|
|
|
is_mine = self.is_mine_id(event.sender)
|
|
|
|
if not is_mine and send_on_behalf_of is None:
|
2021-04-28 10:04:38 +00:00
|
|
|
return
|
2019-03-13 20:02:56 +00:00
|
|
|
|
2019-06-17 17:04:42 +00:00
|
|
|
if not event.internal_metadata.should_proactively_send():
|
2021-04-28 10:04:38 +00:00
|
|
|
return
|
2019-06-17 17:04:42 +00:00
|
|
|
|
2021-07-15 10:02:43 +00:00
|
|
|
destinations: Optional[Set[str]] = None
|
2021-01-26 13:57:31 +00:00
|
|
|
if not event.prev_event_ids():
|
|
|
|
# If there are no prev event IDs then the state is empty
|
|
|
|
# and so no remote servers in the room
|
|
|
|
destinations = set()
|
|
|
|
else:
|
|
|
|
# We check the external cache for the destinations, which is
|
|
|
|
# stored per state group.
|
|
|
|
|
|
|
|
sg = await self._external_cache.get(
|
|
|
|
"event_to_prev_state_group", event.event_id
|
2019-03-13 20:02:56 +00:00
|
|
|
)
|
2021-01-26 13:57:31 +00:00
|
|
|
if sg:
|
|
|
|
destinations = await self._external_cache.get(
|
|
|
|
"get_joined_hosts", str(sg)
|
|
|
|
)
|
|
|
|
|
|
|
|
if destinations is None:
|
|
|
|
try:
|
|
|
|
# Get the state from before the event.
|
|
|
|
# We need to make sure that this is the state from before
|
|
|
|
# the event and not from after it.
|
|
|
|
# Otherwise if the last member on a server in a room is
|
|
|
|
# banned then it won't receive the event because it won't
|
|
|
|
# be in the room after the ban.
|
|
|
|
destinations = await self.state.get_hosts_in_room_at_events(
|
|
|
|
event.room_id, event_ids=event.prev_event_ids()
|
|
|
|
)
|
|
|
|
except Exception:
|
|
|
|
logger.exception(
|
|
|
|
"Failed to calculate hosts in room for event: %s",
|
|
|
|
event.event_id,
|
|
|
|
)
|
2021-04-28 10:04:38 +00:00
|
|
|
return
|
2019-03-13 20:02:56 +00:00
|
|
|
|
2020-07-10 17:26:36 +00:00
|
|
|
destinations = {
|
|
|
|
d
|
|
|
|
for d in destinations
|
2020-07-16 13:06:28 +00:00
|
|
|
if self._federation_shard_config.should_handle(
|
2020-07-10 17:26:36 +00:00
|
|
|
self._instance_name, d
|
|
|
|
)
|
|
|
|
}
|
2019-03-13 20:02:56 +00:00
|
|
|
|
|
|
|
if send_on_behalf_of is not None:
|
|
|
|
# If we are sending the event on behalf of another server
|
|
|
|
# then it already has the event and there is no reason to
|
|
|
|
# send the event to it.
|
|
|
|
destinations.discard(send_on_behalf_of)
|
|
|
|
|
2021-04-28 10:04:38 +00:00
|
|
|
logger.debug("Sending %s to %r", event, destinations)
|
|
|
|
|
2020-06-30 15:58:06 +00:00
|
|
|
if destinations:
|
2021-04-28 10:04:38 +00:00
|
|
|
await self._send_pdu(event, destinations)
|
|
|
|
|
2020-06-30 15:58:06 +00:00
|
|
|
now = self.clock.time_msec()
|
|
|
|
ts = await self.store.get_received_ts(event.event_id)
|
2021-09-10 16:03:18 +00:00
|
|
|
assert ts is not None
|
2020-06-30 15:58:06 +00:00
|
|
|
synapse.metrics.event_processing_lag_by_event.labels(
|
|
|
|
"federation_sender"
|
2020-07-01 14:23:58 +00:00
|
|
|
).observe((now - ts) / 1000)
|
2019-03-13 20:02:56 +00:00
|
|
|
|
2021-04-28 10:04:38 +00:00
|
|
|
async def handle_room_events(events: Iterable[EventBase]) -> None:
|
|
|
|
with Measure(self.clock, "handle_room_events"):
|
|
|
|
for event in events:
|
|
|
|
await handle_event(event)
|
|
|
|
|
2021-07-15 10:02:43 +00:00
|
|
|
events_by_room: Dict[str, List[EventBase]] = {}
|
2021-04-28 10:04:38 +00:00
|
|
|
for event in events:
|
|
|
|
events_by_room.setdefault(event.room_id, []).append(event)
|
|
|
|
|
|
|
|
await make_deferred_yieldable(
|
|
|
|
defer.gatherResults(
|
|
|
|
[
|
|
|
|
run_in_background(handle_room_events, evs)
|
|
|
|
for evs in events_by_room.values()
|
|
|
|
],
|
|
|
|
consumeErrors=True,
|
|
|
|
)
|
|
|
|
)
|
2019-03-13 20:02:56 +00:00
|
|
|
|
2020-02-07 13:56:38 +00:00
|
|
|
await self.store.update_federation_out_pos("events", next_token)
|
2019-06-20 09:32:02 +00:00
|
|
|
|
2019-03-13 20:02:56 +00:00
|
|
|
if events:
|
|
|
|
now = self.clock.time_msec()
|
2020-02-07 13:56:38 +00:00
|
|
|
ts = await self.store.get_received_ts(events[-1].event_id)
|
2021-09-10 16:03:18 +00:00
|
|
|
assert ts is not None
|
2019-03-13 20:02:56 +00:00
|
|
|
|
|
|
|
synapse.metrics.event_processing_lag.labels(
|
|
|
|
"federation_sender"
|
|
|
|
).set(now - ts)
|
|
|
|
synapse.metrics.event_processing_last_ts.labels(
|
|
|
|
"federation_sender"
|
|
|
|
).set(ts)
|
|
|
|
|
|
|
|
events_processed_counter.inc(len(events))
|
|
|
|
|
|
|
|
event_processing_loop_room_count.labels("federation_sender").inc(
|
2021-04-28 10:04:38 +00:00
|
|
|
len(events_by_room)
|
2019-03-13 20:02:56 +00:00
|
|
|
)
|
|
|
|
|
|
|
|
event_processing_loop_counter.labels("federation_sender").inc()
|
|
|
|
|
|
|
|
synapse.metrics.event_processing_positions.labels(
|
|
|
|
"federation_sender"
|
|
|
|
).set(next_token)
|
|
|
|
|
|
|
|
finally:
|
|
|
|
self._is_processing = False
|
|
|
|
|
2021-04-28 10:04:38 +00:00
|
|
|
async def _send_pdu(self, pdu: EventBase, destinations: Iterable[str]) -> None:
|
|
|
|
# We loop through all destinations to see whether we already have
|
|
|
|
# a transaction in progress. If we do, stick it in the pending_pdus
|
|
|
|
# table and we'll get back to it later.
|
2021-04-14 16:19:02 +00:00
|
|
|
|
2021-04-28 10:04:38 +00:00
|
|
|
destinations = set(destinations)
|
|
|
|
destinations.discard(self.server_name)
|
|
|
|
logger.debug("Sending to: %s", str(destinations))
|
2019-03-13 20:02:56 +00:00
|
|
|
|
2021-04-28 10:04:38 +00:00
|
|
|
if not destinations:
|
|
|
|
return
|
2020-10-05 13:43:14 +00:00
|
|
|
|
2021-04-28 10:04:38 +00:00
|
|
|
sent_pdus_destination_dist_total.inc(len(destinations))
|
|
|
|
sent_pdus_destination_dist_count.inc()
|
2021-04-14 16:19:02 +00:00
|
|
|
|
2021-04-28 10:04:38 +00:00
|
|
|
assert pdu.internal_metadata.stream_ordering
|
2021-04-14 16:19:02 +00:00
|
|
|
|
2021-04-28 10:04:38 +00:00
|
|
|
# track the fact that we have a PDU for these destinations,
|
|
|
|
# to allow us to perform catch-up later on if the remote is unreachable
|
|
|
|
# for a while.
|
|
|
|
await self.store.store_destination_rooms_entries(
|
|
|
|
destinations,
|
|
|
|
pdu.room_id,
|
|
|
|
pdu.internal_metadata.stream_ordering,
|
2020-09-04 11:22:23 +00:00
|
|
|
)
|
|
|
|
|
2021-04-28 10:04:38 +00:00
|
|
|
for destination in destinations:
|
|
|
|
self._get_per_destination_queue(destination).send_pdu(pdu)
|
2019-03-13 20:02:56 +00:00
|
|
|
|
2020-07-30 12:01:33 +00:00
|
|
|
async def send_read_receipt(self, receipt: ReadReceipt) -> None:
|
2019-03-13 20:02:56 +00:00
|
|
|
"""Send a RR to any other servers in the room
|
|
|
|
|
|
|
|
Args:
|
2020-02-07 13:56:38 +00:00
|
|
|
receipt: receipt to be sent
|
2019-03-13 20:02:56 +00:00
|
|
|
"""
|
2019-03-20 16:02:25 +00:00
|
|
|
|
|
|
|
# Some background on the rate-limiting going on here.
|
|
|
|
#
|
|
|
|
# It turns out that if we attempt to send out RRs as soon as we get them from
|
|
|
|
# a client, then we end up trying to do several hundred Hz of federation
|
|
|
|
# transactions. (The number of transactions scales as O(N^2) on the size of a
|
|
|
|
# room, since in a large room we have both more RRs coming in, and more servers
|
|
|
|
# to send them to.)
|
|
|
|
#
|
|
|
|
# This leads to a lot of CPU load, and we end up getting behind. The solution
|
|
|
|
# currently adopted is as follows:
|
|
|
|
#
|
|
|
|
# The first receipt in a given room is sent out immediately, at time T0. Any
|
|
|
|
# further receipts are, in theory, batched up for N seconds, where N is calculated
|
|
|
|
# based on the number of servers in the room to achieve a transaction frequency
|
|
|
|
# of around 50Hz. So, for example, if there were 100 servers in the room, then
|
|
|
|
# N would be 100 / 50Hz = 2 seconds.
|
|
|
|
#
|
|
|
|
# Then, after T+N, we flush out any receipts that have accumulated, and restart
|
|
|
|
# the timer to flush out more receipts at T+2N, etc. If no receipts accumulate,
|
|
|
|
# we stop the cycle and go back to the start.
|
|
|
|
#
|
|
|
|
# However, in practice, it is often possible to flush out receipts earlier: in
|
|
|
|
# particular, if we are sending a transaction to a given server anyway (for
|
|
|
|
# example, because we have a PDU or a RR in another room to send), then we may
|
|
|
|
# as well send out all of the pending RRs for that server. So it may be that
|
|
|
|
# by the time we get to T+N, we don't actually have any RRs left to send out.
|
|
|
|
# Nevertheless we continue to buffer up RRs for the room in question until we
|
|
|
|
# reach the point that no RRs arrive between timer ticks.
|
|
|
|
#
|
|
|
|
# For even more background, see https://github.com/matrix-org/synapse/issues/4730.
|
|
|
|
|
|
|
|
room_id = receipt.room_id
|
|
|
|
|
2019-03-13 20:02:56 +00:00
|
|
|
# Work out which remote servers should be poked and poke them.
|
2020-08-24 18:25:27 +00:00
|
|
|
domains_set = await self.state.get_current_hosts_in_room(room_id)
|
2020-07-10 17:26:36 +00:00
|
|
|
domains = [
|
|
|
|
d
|
2020-08-24 18:25:27 +00:00
|
|
|
for d in domains_set
|
2020-07-10 17:26:36 +00:00
|
|
|
if d != self.server_name
|
2020-07-16 13:06:28 +00:00
|
|
|
and self._federation_shard_config.should_handle(self._instance_name, d)
|
2020-07-10 17:26:36 +00:00
|
|
|
]
|
2019-03-13 20:02:56 +00:00
|
|
|
if not domains:
|
|
|
|
return
|
|
|
|
|
2019-03-20 16:02:25 +00:00
|
|
|
queues_pending_flush = self._queues_awaiting_rr_flush_by_room.get(room_id)
|
2019-03-13 20:02:56 +00:00
|
|
|
|
2019-03-20 16:02:25 +00:00
|
|
|
# if there is no flush yet scheduled, we will send out these receipts with
|
|
|
|
# immediate flushes, and schedule the next flush for this room.
|
|
|
|
if queues_pending_flush is not None:
|
|
|
|
logger.debug("Queuing receipt for: %r", domains)
|
|
|
|
else:
|
|
|
|
logger.debug("Sending receipt to: %r", domains)
|
|
|
|
self._schedule_rr_flush_for_room(room_id, len(domains))
|
2019-03-13 20:02:56 +00:00
|
|
|
|
|
|
|
for domain in domains:
|
2019-03-20 16:02:25 +00:00
|
|
|
queue = self._get_per_destination_queue(domain)
|
|
|
|
queue.queue_read_receipt(receipt)
|
|
|
|
|
|
|
|
# if there is already a RR flush pending for this room, then make sure this
|
|
|
|
# destination is registered for the flush
|
|
|
|
if queues_pending_flush is not None:
|
|
|
|
queues_pending_flush.add(queue)
|
|
|
|
else:
|
|
|
|
queue.flush_read_receipts_for_room(room_id)
|
|
|
|
|
2020-02-07 13:56:38 +00:00
|
|
|
def _schedule_rr_flush_for_room(self, room_id: str, n_domains: int) -> None:
|
2019-03-20 16:02:25 +00:00
|
|
|
# that is going to cause approximately len(domains) transactions, so now back
|
|
|
|
# off for that multiplied by RR_TXN_INTERVAL_PER_ROOM
|
|
|
|
backoff_ms = self._rr_txn_interval_per_room_ms * n_domains
|
|
|
|
|
|
|
|
logger.debug("Scheduling RR flush in %s in %d ms", room_id, backoff_ms)
|
|
|
|
self.clock.call_later(backoff_ms, self._flush_rrs_for_room, room_id)
|
|
|
|
self._queues_awaiting_rr_flush_by_room[room_id] = set()
|
|
|
|
|
2020-02-07 13:56:38 +00:00
|
|
|
def _flush_rrs_for_room(self, room_id: str) -> None:
|
2019-03-20 16:02:25 +00:00
|
|
|
queues = self._queues_awaiting_rr_flush_by_room.pop(room_id)
|
|
|
|
logger.debug("Flushing RRs in %s to %s", room_id, queues)
|
|
|
|
|
|
|
|
if not queues:
|
|
|
|
# no more RRs arrived for this room; we are done.
|
|
|
|
return
|
|
|
|
|
|
|
|
# schedule the next flush
|
|
|
|
self._schedule_rr_flush_for_room(room_id, len(queues))
|
|
|
|
|
|
|
|
for queue in queues:
|
|
|
|
queue.flush_read_receipts_for_room(room_id)
|
2019-03-13 20:02:56 +00:00
|
|
|
|
2020-02-07 13:56:38 +00:00
|
|
|
def send_presence_to_destinations(
|
Be smarter about which hosts to send presence to when processing room joins (#9402)
This PR attempts to eliminate unnecessary presence sending work when your local server joins a room, or when a remote server joins a room your server is participating in by processing state deltas in chunks rather than individually.
---
When your server joins a room for the first time, it requests the historical state as well. This chunk of new state is passed to the presence handler which, after filtering that state down to only membership joins, will send presence updates to homeservers for each join processed.
It turns out that we were being a bit naive and processing each event individually, and sending out presence updates for every one of those joins. Even if many different joins were users on the same server (hello IRC bridges), we'd send presence to that same homeserver for every remote user join we saw.
This PR attempts to deduplicate all of that by processing the entire batch of state deltas at once, instead of only doing each join individually. We process the joins and note down which servers need which presence:
* If it was a local user join, send that user's latest presence to all servers in the room
* If it was a remote user join, send the presence for all local users in the room to that homeserver
We deduplicate by inserting all of those pending updates into a dictionary of the form:
```
{
server_name1: {presence_update1, ...},
server_name2: {presence_update1, presence_update2, ...}
}
```
Only after building this dict do we then start sending out presence updates.
2021-02-19 11:37:29 +00:00
|
|
|
self, states: Iterable[UserPresenceState], destinations: Iterable[str]
|
2020-02-07 13:56:38 +00:00
|
|
|
) -> None:
|
2019-03-26 12:45:22 +00:00
|
|
|
"""Send the given presence states to the given destinations.
|
|
|
|
destinations (list[str])
|
|
|
|
"""
|
|
|
|
|
|
|
|
if not states or not self.hs.config.use_presence:
|
|
|
|
# No-op if presence is disabled.
|
|
|
|
return
|
|
|
|
|
2021-04-20 12:37:54 +00:00
|
|
|
# Ensure we only send out presence states for local users.
|
|
|
|
for state in states:
|
|
|
|
assert self.is_mine_id(state.user_id)
|
|
|
|
|
2019-03-26 12:45:22 +00:00
|
|
|
for destination in destinations:
|
|
|
|
if destination == self.server_name:
|
|
|
|
continue
|
2020-07-16 13:06:28 +00:00
|
|
|
if not self._federation_shard_config.should_handle(
|
2020-07-10 17:26:36 +00:00
|
|
|
self._instance_name, destination
|
|
|
|
):
|
|
|
|
continue
|
2021-07-15 10:52:56 +00:00
|
|
|
|
|
|
|
self._get_per_destination_queue(destination).send_presence(
|
|
|
|
states, start_loop=False
|
|
|
|
)
|
|
|
|
|
|
|
|
self._presence_queue.add_to_queue(destination)
|
2019-03-26 12:45:22 +00:00
|
|
|
|
2020-02-07 13:56:38 +00:00
|
|
|
def build_and_send_edu(
|
|
|
|
self,
|
|
|
|
destination: str,
|
|
|
|
edu_type: str,
|
2021-03-29 15:43:20 +00:00
|
|
|
content: JsonDict,
|
2020-02-07 13:56:38 +00:00
|
|
|
key: Optional[Hashable] = None,
|
2021-03-29 15:43:20 +00:00
|
|
|
) -> None:
|
2019-03-13 20:02:56 +00:00
|
|
|
"""Construct an Edu object, and queue it for sending
|
|
|
|
|
|
|
|
Args:
|
2020-02-07 13:56:38 +00:00
|
|
|
destination: name of server to send to
|
|
|
|
edu_type: type of EDU to send
|
|
|
|
content: content of EDU
|
|
|
|
key: clobbering key for this edu
|
2019-03-13 20:02:56 +00:00
|
|
|
"""
|
|
|
|
if destination == self.server_name:
|
|
|
|
logger.info("Not sending EDU to ourselves")
|
|
|
|
return
|
|
|
|
|
2020-07-16 13:06:28 +00:00
|
|
|
if not self._federation_shard_config.should_handle(
|
2020-07-10 17:26:36 +00:00
|
|
|
self._instance_name, destination
|
|
|
|
):
|
|
|
|
return
|
|
|
|
|
2019-03-13 20:02:56 +00:00
|
|
|
edu = Edu(
|
|
|
|
origin=self.server_name,
|
|
|
|
destination=destination,
|
|
|
|
edu_type=edu_type,
|
|
|
|
content=content,
|
|
|
|
)
|
|
|
|
|
|
|
|
self.send_edu(edu, key)
|
|
|
|
|
2021-03-29 15:43:20 +00:00
|
|
|
def send_edu(self, edu: Edu, key: Optional[Hashable]) -> None:
|
2019-03-13 20:02:56 +00:00
|
|
|
"""Queue an EDU for sending
|
|
|
|
|
|
|
|
Args:
|
2020-02-07 13:56:38 +00:00
|
|
|
edu: edu to send
|
|
|
|
key: clobbering key for this edu
|
2019-03-13 20:02:56 +00:00
|
|
|
"""
|
2020-07-16 13:06:28 +00:00
|
|
|
if not self._federation_shard_config.should_handle(
|
2020-07-10 17:26:36 +00:00
|
|
|
self._instance_name, edu.destination
|
|
|
|
):
|
|
|
|
return
|
|
|
|
|
2019-03-13 20:02:56 +00:00
|
|
|
queue = self._get_per_destination_queue(edu.destination)
|
|
|
|
if key:
|
|
|
|
queue.send_keyed_edu(edu, key)
|
|
|
|
else:
|
|
|
|
queue.send_edu(edu)
|
|
|
|
|
2021-03-29 15:43:20 +00:00
|
|
|
def send_device_messages(self, destination: str) -> None:
|
2019-03-13 20:02:56 +00:00
|
|
|
if destination == self.server_name:
|
2020-01-17 10:27:19 +00:00
|
|
|
logger.warning("Not sending device update to ourselves")
|
|
|
|
return
|
|
|
|
|
2020-07-16 13:06:28 +00:00
|
|
|
if not self._federation_shard_config.should_handle(
|
2020-07-10 17:26:36 +00:00
|
|
|
self._instance_name, destination
|
|
|
|
):
|
|
|
|
return
|
|
|
|
|
2020-01-17 10:27:19 +00:00
|
|
|
self._get_per_destination_queue(destination).attempt_new_transaction()
|
|
|
|
|
2021-03-29 15:43:20 +00:00
|
|
|
def wake_destination(self, destination: str) -> None:
|
2020-01-17 10:27:19 +00:00
|
|
|
"""Called when we want to retry sending transactions to a remote.
|
|
|
|
|
|
|
|
This is mainly useful if the remote server has been down and we think it
|
|
|
|
might have come back.
|
|
|
|
"""
|
|
|
|
|
|
|
|
if destination == self.server_name:
|
|
|
|
logger.warning("Not waking up ourselves")
|
2019-03-13 20:02:56 +00:00
|
|
|
return
|
|
|
|
|
2020-07-16 13:06:28 +00:00
|
|
|
if not self._federation_shard_config.should_handle(
|
2020-07-10 17:26:36 +00:00
|
|
|
self._instance_name, destination
|
|
|
|
):
|
|
|
|
return
|
|
|
|
|
2019-03-13 20:02:56 +00:00
|
|
|
self._get_per_destination_queue(destination).attempt_new_transaction()
|
|
|
|
|
2020-05-05 13:15:57 +00:00
|
|
|
@staticmethod
|
|
|
|
def get_current_token() -> int:
|
2020-03-25 14:54:01 +00:00
|
|
|
# Dummy implementation for case where federation sender isn't offloaded
|
|
|
|
# to a worker.
|
2019-03-13 20:02:56 +00:00
|
|
|
return 0
|
2020-03-25 14:54:01 +00:00
|
|
|
|
2021-03-29 15:43:20 +00:00
|
|
|
def federation_ack(self, instance_name: str, token: int) -> None:
|
|
|
|
# It is not expected that this gets called on FederationSender.
|
|
|
|
raise NotImplementedError()
|
|
|
|
|
2020-05-05 13:15:57 +00:00
|
|
|
@staticmethod
|
2020-03-25 14:54:01 +00:00
|
|
|
async def get_replication_rows(
|
2020-05-05 13:15:57 +00:00
|
|
|
instance_name: str, from_token: int, to_token: int, target_row_count: int
|
|
|
|
) -> Tuple[List[Tuple[int, Tuple]], int, bool]:
|
2020-03-25 14:54:01 +00:00
|
|
|
# Dummy implementation for case where federation sender isn't offloaded
|
|
|
|
# to a worker.
|
2020-05-05 13:15:57 +00:00
|
|
|
return [], 0, False
|
2020-09-18 13:59:13 +00:00
|
|
|
|
2021-03-29 15:43:20 +00:00
|
|
|
async def _wake_destinations_needing_catchup(self) -> None:
|
2020-09-18 13:59:13 +00:00
|
|
|
"""
|
|
|
|
Wakes up destinations that need catch-up and are not currently being
|
|
|
|
backed off from.
|
|
|
|
|
|
|
|
In order to reduce load spikes, adds a delay between each destination.
|
|
|
|
"""
|
|
|
|
|
2021-07-15 10:02:43 +00:00
|
|
|
last_processed: Optional[str] = None
|
2020-09-18 13:59:13 +00:00
|
|
|
|
|
|
|
while True:
|
|
|
|
destinations_to_wake = (
|
|
|
|
await self.store.get_catch_up_outstanding_destinations(last_processed)
|
|
|
|
)
|
|
|
|
|
|
|
|
if not destinations_to_wake:
|
|
|
|
# finished waking all destinations!
|
|
|
|
self._catchup_after_startup_timer = None
|
|
|
|
break
|
|
|
|
|
2021-04-08 16:30:01 +00:00
|
|
|
last_processed = destinations_to_wake[-1]
|
|
|
|
|
2020-09-18 13:59:13 +00:00
|
|
|
destinations_to_wake = [
|
|
|
|
d
|
|
|
|
for d in destinations_to_wake
|
|
|
|
if self._federation_shard_config.should_handle(self._instance_name, d)
|
|
|
|
]
|
|
|
|
|
2021-04-08 16:30:01 +00:00
|
|
|
for destination in destinations_to_wake:
|
2020-09-18 13:59:13 +00:00
|
|
|
logger.info(
|
|
|
|
"Destination %s has outstanding catch-up, waking up.",
|
|
|
|
last_processed,
|
|
|
|
)
|
2021-04-08 16:30:01 +00:00
|
|
|
self.wake_destination(destination)
|
2020-09-18 13:59:13 +00:00
|
|
|
await self.clock.sleep(CATCH_UP_STARTUP_INTERVAL_SEC)
|