synapse-old/synapse/util/ratelimitutils.py

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

196 lines
6.8 KiB
Python
Raw Normal View History

2016-01-06 21:26:29 -07:00
# Copyright 2015, 2016 OpenMarket 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.
2018-07-09 00:09:20 -06:00
import collections
import contextlib
import logging
2021-09-10 10:03:18 -06:00
import typing
from typing import Any, DefaultDict, Iterator, List, Set
2018-07-09 00:09:20 -06:00
from twisted.internet import defer
from synapse.api.errors import LimitExceededError
2021-09-10 10:03:18 -06:00
from synapse.config.ratelimiting import FederationRateLimitConfig
from synapse.logging.context import (
2018-05-03 04:16:36 -06:00
PreserveLoggingContext,
2018-07-09 00:09:20 -06:00
make_deferred_yieldable,
run_in_background,
2018-05-03 04:16:36 -06:00
)
2021-09-10 10:03:18 -06:00
from synapse.util import Clock
if typing.TYPE_CHECKING:
from contextlib import _GeneratorContextManager
logger = logging.getLogger(__name__)
2020-09-04 04:54:56 -06:00
class FederationRateLimiter:
2021-09-10 10:03:18 -06:00
def __init__(self, clock: Clock, config: FederationRateLimitConfig):
def new_limiter() -> "_PerHostRatelimiter":
return _PerHostRatelimiter(clock=clock, config=config)
2021-09-10 10:03:18 -06:00
self.ratelimiters: DefaultDict[
str, "_PerHostRatelimiter"
] = collections.defaultdict(new_limiter)
2021-09-10 10:03:18 -06:00
def ratelimit(self, host: str) -> "_GeneratorContextManager[defer.Deferred[None]]":
"""Used to ratelimit an incoming request from a given host
Example usage:
with rate_limiter.ratelimit(origin) as wait_deferred:
yield wait_deferred
# Handle request ...
Args:
host (str): Origin of incoming request.
Returns:
context manager which returns a deferred.
"""
return self.ratelimiters[host].ratelimit()
2020-09-04 04:54:56 -06:00
class _PerHostRatelimiter:
2021-09-10 10:03:18 -06:00
def __init__(self, clock: Clock, config: FederationRateLimitConfig):
"""
Args:
2021-09-10 10:03:18 -06:00
clock
config
"""
self.clock = clock
self.window_size = config.window_size
self.sleep_limit = config.sleep_limit
self.sleep_sec = config.sleep_delay / 1000.0
self.reject_limit = config.reject_limit
self.concurrent_requests = config.concurrent
# request_id objects for requests which have been slept
2021-09-10 10:03:18 -06:00
self.sleeping_requests: Set[object] = set()
# map from request_id object to Deferred for requests which are ready
# for processing but have been queued
2021-09-10 10:03:18 -06:00
self.ready_request_queue: collections.OrderedDict[
object, defer.Deferred[None]
] = collections.OrderedDict()
# request id objects for requests which are in progress
2021-09-10 10:03:18 -06:00
self.current_processing: Set[object] = set()
# times at which we have recently (within the last window_size ms)
# received requests.
2021-09-10 10:03:18 -06:00
self.request_times: List[int] = []
@contextlib.contextmanager
2021-09-10 10:03:18 -06:00
def ratelimit(self) -> "Iterator[defer.Deferred[None]]":
# `contextlib.contextmanager` takes a generator and turns it into a
# context manager. The generator should only yield once with a value
# to be returned by manager.
# Exceptions will be reraised at the yield.
request_id = object()
ret = self._on_enter(request_id)
try:
yield ret
finally:
self._on_exit(request_id)
2021-09-10 10:03:18 -06:00
def _on_enter(self, request_id: object) -> "defer.Deferred[None]":
time_now = self.clock.time_msec()
# remove any entries from request_times which aren't within the window
self.request_times[:] = [
r for r in self.request_times if time_now - r < self.window_size
]
# reject the request if we already have too many queued up (either
# sleeping or in the ready queue).
queue_size = len(self.ready_request_queue) + len(self.sleeping_requests)
if queue_size > self.reject_limit:
raise LimitExceededError(
retry_after_ms=int(self.window_size / self.sleep_limit)
)
self.request_times.append(time_now)
2021-09-10 10:03:18 -06:00
def queue_request() -> "defer.Deferred[None]":
if len(self.current_processing) >= self.concurrent_requests:
2021-09-10 10:03:18 -06:00
queue_defer: defer.Deferred[None] = defer.Deferred()
self.ready_request_queue[request_id] = queue_defer
logger.info(
"Ratelimiter: queueing request (queue now %i items)",
len(self.ready_request_queue),
)
return queue_defer
else:
return defer.succeed(None)
logger.debug(
"Ratelimit [%s]: len(self.request_times)=%d",
id(request_id),
len(self.request_times),
)
if len(self.request_times) > self.sleep_limit:
logger.debug("Ratelimiter: sleeping request for %f sec", self.sleep_sec)
ret_defer = run_in_background(self.clock.sleep, self.sleep_sec)
self.sleeping_requests.add(request_id)
2021-09-10 10:03:18 -06:00
def on_wait_finished(_: Any) -> "defer.Deferred[None]":
logger.debug("Ratelimit [%s]: Finished sleeping", id(request_id))
self.sleeping_requests.discard(request_id)
queue_defer = queue_request()
return queue_defer
ret_defer.addBoth(on_wait_finished)
else:
ret_defer = queue_request()
2021-09-10 10:03:18 -06:00
def on_start(r: object) -> object:
logger.debug("Ratelimit [%s]: Processing req", id(request_id))
self.current_processing.add(request_id)
return r
2021-09-10 10:03:18 -06:00
def on_err(r: object) -> object:
2018-05-03 04:16:36 -06:00
# XXX: why is this necessary? this is called before we start
# processing the request so why would the request be in
# current_processing?
self.current_processing.discard(request_id)
return r
2021-09-10 10:03:18 -06:00
def on_both(r: object) -> object:
# Ensure that we've properly cleaned up.
self.sleeping_requests.discard(request_id)
self.ready_request_queue.pop(request_id, None)
return r
ret_defer.addCallbacks(on_start, on_err)
ret_defer.addBoth(on_both)
2018-05-03 04:16:36 -06:00
return make_deferred_yieldable(ret_defer)
2021-09-10 10:03:18 -06:00
def _on_exit(self, request_id: object) -> None:
logger.debug("Ratelimit [%s]: Processed req", id(request_id))
self.current_processing.discard(request_id)
try:
# start processing the next item on the queue.
_, deferred = self.ready_request_queue.popitem(last=False)
2018-05-03 04:16:36 -06:00
with PreserveLoggingContext():
deferred.callback(None)
except KeyError:
pass