forked-synapse/synapse/federation/sender/per_destination_queue.py
Erik Johnston ac5c221208
Stagger send presence to remotes (#10398)
This is to help with performance, where trying to connect to thousands
of hosts at once can consume a lot of CPU (due to TLS etc).

Co-authored-by: Brendan Abolivier <babolivier@matrix.org>
2021-07-15 11:52:56 +01:00

738 lines
29 KiB
Python
Raw Blame History

This file contains invisible Unicode characters

This file contains invisible Unicode characters that are indistinguishable to humans but may be processed differently by a computer. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

# Copyright 2014-2016 OpenMarket Ltd
# 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.
import datetime
import logging
from typing import TYPE_CHECKING, Dict, Hashable, Iterable, List, Optional, Tuple
import attr
from prometheus_client import Counter
from synapse.api.errors import (
FederationDeniedError,
HttpResponseException,
RequestSendFailed,
)
from synapse.api.presence import UserPresenceState
from synapse.events import EventBase
from synapse.federation.units import Edu
from synapse.handlers.presence import format_user_presence_state
from synapse.logging import issue9533_logger
from synapse.logging.opentracing import SynapseTags, set_tag
from synapse.metrics import sent_transactions_counter
from synapse.metrics.background_process_metrics import run_as_background_process
from synapse.types import ReadReceipt
from synapse.util.retryutils import NotRetryingDestination, get_retry_limiter
if TYPE_CHECKING:
import synapse.server
# This is defined in the Matrix spec and enforced by the receiver.
MAX_EDUS_PER_TRANSACTION = 100
logger = logging.getLogger(__name__)
sent_edus_counter = Counter(
"synapse_federation_client_sent_edus", "Total number of EDUs successfully sent"
)
sent_edus_by_type = Counter(
"synapse_federation_client_sent_edus_by_type",
"Number of sent EDUs successfully sent, by event type",
["type"],
)
class PerDestinationQueue:
"""
Manages the per-destination transmission queues.
Args:
hs
transaction_sender
destination: the server_name of the destination that we are managing
transmission for.
"""
def __init__(
self,
hs: "synapse.server.HomeServer",
transaction_manager: "synapse.federation.sender.TransactionManager",
destination: str,
):
self._server_name = hs.hostname
self._clock = hs.get_clock()
self._store = hs.get_datastore()
self._transaction_manager = transaction_manager
self._instance_name = hs.get_instance_name()
self._federation_shard_config = hs.config.worker.federation_shard_config
self._state = hs.get_state_handler()
self._should_send_on_this_instance = True
if not self._federation_shard_config.should_handle(
self._instance_name, destination
):
# We don't raise an exception here to avoid taking out any other
# processing. We have a guard in `attempt_new_transaction` that
# ensure we don't start sending stuff.
logger.error(
"Create a per destination queue for %s on wrong worker",
destination,
)
self._should_send_on_this_instance = False
self._destination = destination
self.transmission_loop_running = False
# Flag to signal to any running transmission loop that there is new data
# queued up to be sent.
self._new_data_to_send = False
# True whilst we are sending events that the remote homeserver missed
# because it was unreachable. We start in this state so we can perform
# catch-up at startup.
# New events will only be sent once this is finished, at which point
# _catching_up is flipped to False.
self._catching_up: bool = True
# The stream_ordering of the most recent PDU that was discarded due to
# being in catch-up mode.
self._catchup_last_skipped: int = 0
# Cache of the last successfully-transmitted stream ordering for this
# destination (we are the only updater so this is safe)
self._last_successful_stream_ordering: Optional[int] = None
# a queue of pending PDUs
self._pending_pdus: List[EventBase] = []
# XXX this is never actually used: see
# https://github.com/matrix-org/synapse/issues/7549
self._pending_edus: List[Edu] = []
# Pending EDUs by their "key". Keyed EDUs are EDUs that get clobbered
# based on their key (e.g. typing events by room_id)
# Map of (edu_type, key) -> Edu
self._pending_edus_keyed: Dict[Tuple[str, Hashable], Edu] = {}
# Map of user_id -> UserPresenceState of pending presence to be sent to this
# destination
self._pending_presence: Dict[str, UserPresenceState] = {}
# room_id -> receipt_type -> user_id -> receipt_dict
self._pending_rrs: Dict[str, Dict[str, Dict[str, dict]]] = {}
self._rrs_pending_flush = False
# stream_id of last successfully sent to-device message.
# NB: may be a long or an int.
self._last_device_stream_id = 0
# stream_id of last successfully sent device list update.
self._last_device_list_stream_id = 0
def __str__(self) -> str:
return "PerDestinationQueue[%s]" % self._destination
def pending_pdu_count(self) -> int:
return len(self._pending_pdus)
def pending_edu_count(self) -> int:
return (
len(self._pending_edus)
+ len(self._pending_presence)
+ len(self._pending_edus_keyed)
)
def send_pdu(self, pdu: EventBase) -> None:
"""Add a PDU to the queue, and start the transmission loop if necessary
Args:
pdu: pdu to send
"""
if not self._catching_up or self._last_successful_stream_ordering is None:
# only enqueue the PDU if we are not catching up (False) or do not
# yet know if we have anything to catch up (None)
self._pending_pdus.append(pdu)
else:
assert pdu.internal_metadata.stream_ordering
self._catchup_last_skipped = pdu.internal_metadata.stream_ordering
self.attempt_new_transaction()
def send_presence(
self, states: Iterable[UserPresenceState], start_loop: bool = True
) -> None:
"""Add presence updates to the queue.
Args:
states: Presence updates to send
start_loop: Whether to start the transmission loop if not already
running.
Args:
states: presence to send
"""
self._pending_presence.update({state.user_id: state for state in states})
self._new_data_to_send = True
if start_loop:
self.attempt_new_transaction()
def queue_read_receipt(self, receipt: ReadReceipt) -> None:
"""Add a RR to the list to be sent. Doesn't start the transmission loop yet
(see flush_read_receipts_for_room)
Args:
receipt: receipt to be queued
"""
self._pending_rrs.setdefault(receipt.room_id, {}).setdefault(
receipt.receipt_type, {}
)[receipt.user_id] = {"event_ids": receipt.event_ids, "data": receipt.data}
def flush_read_receipts_for_room(self, room_id: str) -> None:
# if we don't have any read-receipts for this room, it may be that we've already
# sent them out, so we don't need to flush.
if room_id not in self._pending_rrs:
return
self._rrs_pending_flush = True
self.attempt_new_transaction()
def send_keyed_edu(self, edu: Edu, key: Hashable) -> None:
self._pending_edus_keyed[(edu.edu_type, key)] = edu
self.attempt_new_transaction()
def send_edu(self, edu) -> None:
self._pending_edus.append(edu)
self.attempt_new_transaction()
def attempt_new_transaction(self) -> None:
"""Try to start a new transaction to this destination
If there is already a transaction in progress to this destination,
returns immediately. Otherwise kicks off the process of sending a
transaction in the background.
"""
# Mark that we (may) have new things to send, so that any running
# transmission loop will recheck whether there is stuff to send.
self._new_data_to_send = True
if self.transmission_loop_running:
# XXX: this can get stuck on by a never-ending
# request at which point pending_pdus just keeps growing.
# we need application-layer timeouts of some flavour of these
# requests
logger.debug("TX [%s] Transaction already in progress", self._destination)
return
if not self._should_send_on_this_instance:
# We don't raise an exception here to avoid taking out any other
# processing.
logger.error(
"Trying to start a transaction to %s on wrong worker", self._destination
)
return
logger.debug("TX [%s] Starting transaction loop", self._destination)
run_as_background_process(
"federation_transaction_transmission_loop",
self._transaction_transmission_loop,
)
async def _transaction_transmission_loop(self) -> None:
pending_pdus: List[EventBase] = []
try:
self.transmission_loop_running = True
# This will throw if we wouldn't retry. We do this here so we fail
# quickly, but we will later check this again in the http client,
# hence why we throw the result away.
await get_retry_limiter(self._destination, self._clock, self._store)
if self._catching_up:
# we potentially need to catch-up first
await self._catch_up_transmission_loop()
if self._catching_up:
# not caught up yet
return
pending_pdus = []
while True:
self._new_data_to_send = False
async with _TransactionQueueManager(self) as (
pending_pdus,
pending_edus,
):
if not pending_pdus and not pending_edus:
logger.debug("TX [%s] Nothing to send", self._destination)
# If we've gotten told about new things to send during
# checking for things to send, we try looking again.
# Otherwise new PDUs or EDUs might arrive in the meantime,
# but not get sent because we hold the
# `transmission_loop_running` flag.
if self._new_data_to_send:
continue
else:
return
if pending_pdus:
logger.debug(
"TX [%s] len(pending_pdus_by_dest[dest]) = %d",
self._destination,
len(pending_pdus),
)
await self._transaction_manager.send_new_transaction(
self._destination, pending_pdus, pending_edus
)
sent_transactions_counter.inc()
sent_edus_counter.inc(len(pending_edus))
for edu in pending_edus:
sent_edus_by_type.labels(edu.edu_type).inc()
except NotRetryingDestination as e:
logger.debug(
"TX [%s] not ready for retry yet (next retry at %s) - "
"dropping transaction for now",
self._destination,
datetime.datetime.fromtimestamp(
(e.retry_last_ts + e.retry_interval) / 1000.0
),
)
if e.retry_interval > 60 * 60 * 1000:
# we won't retry for another hour!
# (this suggests a significant outage)
# We drop pending EDUs because otherwise they will
# rack up indefinitely.
# (Dropping PDUs is already performed by `_start_catching_up`.)
# Note that:
# - the EDUs that are being dropped here are those that we can
# afford to drop (specifically, only typing notifications,
# read receipts and presence updates are being dropped here)
# - Other EDUs such as to_device messages are queued with a
# different mechanism
# - this is all volatile state that would be lost if the
# federation sender restarted anyway
# dropping read receipts is a bit sad but should be solved
# through another mechanism, because this is all volatile!
self._pending_edus = []
self._pending_edus_keyed = {}
self._pending_presence = {}
self._pending_rrs = {}
self._start_catching_up()
except FederationDeniedError as e:
logger.info(e)
except HttpResponseException as e:
logger.warning(
"TX [%s] Received %d response to transaction: %s",
self._destination,
e.code,
e,
)
except RequestSendFailed as e:
logger.warning(
"TX [%s] Failed to send transaction: %s", self._destination, e
)
for p in pending_pdus:
logger.info(
"Failed to send event %s to %s", p.event_id, self._destination
)
except Exception:
logger.exception("TX [%s] Failed to send transaction", self._destination)
for p in pending_pdus:
logger.info(
"Failed to send event %s to %s", p.event_id, self._destination
)
finally:
# We want to be *very* sure we clear this after we stop processing
self.transmission_loop_running = False
async def _catch_up_transmission_loop(self) -> None:
first_catch_up_check = self._last_successful_stream_ordering is None
if first_catch_up_check:
# first catchup so get last_successful_stream_ordering from database
self._last_successful_stream_ordering = (
await self._store.get_destination_last_successful_stream_ordering(
self._destination
)
)
if self._last_successful_stream_ordering is None:
# if it's still None, then this means we don't have the information
# in our database ­ we haven't successfully sent a PDU to this server
# (at least since the introduction of the feature tracking
# last_successful_stream_ordering).
# Sadly, this means we can't do anything here as we don't know what
# needs catching up — so catching up is futile; let's stop.
self._catching_up = False
return
# get at most 50 catchup room/PDUs
while True:
event_ids = await self._store.get_catch_up_room_event_ids(
self._destination,
self._last_successful_stream_ordering,
)
if not event_ids:
# No more events to catch up on, but we can't ignore the chance
# of a race condition, so we check that no new events have been
# skipped due to us being in catch-up mode
if self._catchup_last_skipped > self._last_successful_stream_ordering:
# another event has been skipped because we were in catch-up mode
continue
# we are done catching up!
self._catching_up = False
break
if first_catch_up_check:
# as this is our check for needing catch-up, we may have PDUs in
# the queue from before we *knew* we had to do catch-up, so
# clear those out now.
self._start_catching_up()
# fetch the relevant events from the event store
# - redacted behaviour of REDACT is fine, since we only send metadata
# of redacted events to the destination.
# - don't need to worry about rejected events as we do not actively
# forward received events over federation.
catchup_pdus = await self._store.get_events_as_list(event_ids)
if not catchup_pdus:
raise AssertionError(
"No events retrieved when we asked for %r. "
"This should not happen." % event_ids
)
# We send transactions with events from one room only, as its likely
# that the remote will have to do additional processing, which may
# take some time. It's better to give it small amounts of work
# rather than risk the request timing out and repeatedly being
# retried, and not making any progress.
#
# Note: `catchup_pdus` will have exactly one PDU per room.
for pdu in catchup_pdus:
# The PDU from the DB will be the last PDU in the room from
# *this server* that wasn't sent to the remote. However, other
# servers may have sent lots of events since then, and we want
# to try and tell the remote only about the *latest* events in
# the room. This is so that it doesn't get inundated by events
# from various parts of the DAG, which all need to be processed.
#
# Note: this does mean that in large rooms a server coming back
# online will get sent the same events from all the different
# servers, but the remote will correctly deduplicate them and
# handle it only once.
# Step 1, fetch the current extremities
extrems = await self._store.get_prev_events_for_room(pdu.room_id)
if pdu.event_id in extrems:
# If the event is in the extremities, then great! We can just
# use that without having to do further checks.
room_catchup_pdus = [pdu]
else:
# If not, fetch the extremities and figure out which we can
# send.
extrem_events = await self._store.get_events_as_list(extrems)
new_pdus = []
for p in extrem_events:
# We pulled this from the DB, so it'll be non-null
assert p.internal_metadata.stream_ordering
# Filter out events that happened before the remote went
# offline
if (
p.internal_metadata.stream_ordering
< self._last_successful_stream_ordering
):
continue
# Filter out events where the server is not in the room,
# e.g. it may have left/been kicked. *Ideally* we'd pull
# out the kick and send that, but it's a rare edge case
# so we don't bother for now (the server that sent the
# kick should send it out if its online).
hosts = await self._state.get_hosts_in_room_at_events(
p.room_id, [p.event_id]
)
if self._destination not in hosts:
continue
new_pdus.append(p)
# If we've filtered out all the extremities, fall back to
# sending the original event. This should ensure that the
# server gets at least some of missed events (especially if
# the other sending servers are up).
if new_pdus:
room_catchup_pdus = new_pdus
else:
room_catchup_pdus = [pdu]
logger.info(
"Catching up rooms to %s: %r", self._destination, pdu.room_id
)
await self._transaction_manager.send_new_transaction(
self._destination, room_catchup_pdus, []
)
sent_transactions_counter.inc()
# We pulled this from the DB, so it'll be non-null
assert pdu.internal_metadata.stream_ordering
# Note that we mark the last successful stream ordering as that
# from the *original* PDU, rather than the PDU(s) we actually
# send. This is because we use it to mark our position in the
# queue of missed PDUs to process.
self._last_successful_stream_ordering = (
pdu.internal_metadata.stream_ordering
)
await self._store.set_destination_last_successful_stream_ordering(
self._destination, self._last_successful_stream_ordering
)
def _get_rr_edus(self, force_flush: bool) -> Iterable[Edu]:
if not self._pending_rrs:
return
if not force_flush and not self._rrs_pending_flush:
# not yet time for this lot
return
edu = Edu(
origin=self._server_name,
destination=self._destination,
edu_type="m.receipt",
content=self._pending_rrs,
)
self._pending_rrs = {}
self._rrs_pending_flush = False
yield edu
def _pop_pending_edus(self, limit: int) -> List[Edu]:
pending_edus = self._pending_edus
pending_edus, self._pending_edus = pending_edus[:limit], pending_edus[limit:]
return pending_edus
async def _get_device_update_edus(self, limit: int) -> Tuple[List[Edu], int]:
last_device_list = self._last_device_list_stream_id
# Retrieve list of new device updates to send to the destination
now_stream_id, results = await self._store.get_device_updates_by_remote(
self._destination, last_device_list, limit=limit
)
edus = [
Edu(
origin=self._server_name,
destination=self._destination,
edu_type=edu_type,
content=content,
)
for (edu_type, content) in results
]
assert len(edus) <= limit, "get_device_updates_by_remote returned too many EDUs"
return (edus, now_stream_id)
async def _get_to_device_message_edus(self, limit: int) -> Tuple[List[Edu], int]:
last_device_stream_id = self._last_device_stream_id
to_device_stream_id = self._store.get_to_device_stream_token()
contents, stream_id = await self._store.get_new_device_msgs_for_remote(
self._destination, last_device_stream_id, to_device_stream_id, limit
)
for content in contents:
message_id = content.get("message_id")
if not message_id:
continue
set_tag(SynapseTags.TO_DEVICE_MESSAGE_ID, message_id)
edus = [
Edu(
origin=self._server_name,
destination=self._destination,
edu_type="m.direct_to_device",
content=content,
)
for content in contents
]
if edus:
issue9533_logger.debug(
"Sending %i to-device messages to %s, up to stream id %i",
len(edus),
self._destination,
stream_id,
)
return (edus, stream_id)
def _start_catching_up(self) -> None:
"""
Marks this destination as being in catch-up mode.
This throws away the PDU queue.
"""
self._catching_up = True
self._pending_pdus = []
@attr.s(slots=True)
class _TransactionQueueManager:
"""A helper async context manager for pulling stuff off the queues and
tracking what was last successfully sent, etc.
"""
queue = attr.ib(type=PerDestinationQueue)
_device_stream_id = attr.ib(type=Optional[int], default=None)
_device_list_id = attr.ib(type=Optional[int], default=None)
_last_stream_ordering = attr.ib(type=Optional[int], default=None)
_pdus = attr.ib(type=List[EventBase], factory=list)
async def __aenter__(self) -> Tuple[List[EventBase], List[Edu]]:
# First we calculate the EDUs we want to send, if any.
# We start by fetching device related EDUs, i.e device updates and to
# device messages. We have to keep 2 free slots for presence and rr_edus.
limit = MAX_EDUS_PER_TRANSACTION - 2
device_update_edus, dev_list_id = await self.queue._get_device_update_edus(
limit
)
if device_update_edus:
self._device_list_id = dev_list_id
else:
self.queue._last_device_list_stream_id = dev_list_id
limit -= len(device_update_edus)
(
to_device_edus,
device_stream_id,
) = await self.queue._get_to_device_message_edus(limit)
if to_device_edus:
self._device_stream_id = device_stream_id
else:
self.queue._last_device_stream_id = device_stream_id
pending_edus = device_update_edus + to_device_edus
# Now add the read receipt EDU.
pending_edus.extend(self.queue._get_rr_edus(force_flush=False))
# And presence EDU.
if self.queue._pending_presence:
pending_edus.append(
Edu(
origin=self.queue._server_name,
destination=self.queue._destination,
edu_type="m.presence",
content={
"push": [
format_user_presence_state(
presence, self.queue._clock.time_msec()
)
for presence in self.queue._pending_presence.values()
]
},
)
)
self.queue._pending_presence = {}
# Finally add any other types of EDUs if there is room.
pending_edus.extend(
self.queue._pop_pending_edus(MAX_EDUS_PER_TRANSACTION - len(pending_edus))
)
while (
len(pending_edus) < MAX_EDUS_PER_TRANSACTION
and self.queue._pending_edus_keyed
):
_, val = self.queue._pending_edus_keyed.popitem()
pending_edus.append(val)
# Now we look for any PDUs to send, by getting up to 50 PDUs from the
# queue
self._pdus = self.queue._pending_pdus[:50]
if not self._pdus and not pending_edus:
return [], []
# if we've decided to send a transaction anyway, and we have room, we
# may as well send any pending RRs
if len(pending_edus) < MAX_EDUS_PER_TRANSACTION:
pending_edus.extend(self.queue._get_rr_edus(force_flush=True))
if self._pdus:
self._last_stream_ordering = self._pdus[
-1
].internal_metadata.stream_ordering
assert self._last_stream_ordering
return self._pdus, pending_edus
async def __aexit__(self, exc_type, exc, tb):
if exc_type is not None:
# Failed to send transaction, so we bail out.
return
# Successfully sent transactions, so we remove pending PDUs from the queue
if self._pdus:
self.queue._pending_pdus = self.queue._pending_pdus[len(self._pdus) :]
# Succeeded to send the transaction so we record where we have sent up
# to in the various streams
if self._device_stream_id:
await self.queue._store.delete_device_msgs_for_remote(
self.queue._destination, self._device_stream_id
)
self.queue._last_device_stream_id = self._device_stream_id
# also mark the device updates as sent
if self._device_list_id:
logger.info(
"Marking as sent %r %r", self.queue._destination, self._device_list_id
)
await self.queue._store.mark_as_sent_devices_by_remote(
self.queue._destination, self._device_list_id
)
self.queue._last_device_list_stream_id = self._device_list_id
if self._last_stream_ordering:
# we sent some PDUs and it was successful, so update our
# last_successful_stream_ordering in the destinations table.
await self.queue._store.set_destination_last_successful_stream_ordering(
self.queue._destination, self._last_stream_ordering
)