synapse-product/synapse/util/async_helpers.py
Sean Quah 800ba87cc8
Refactor and convert Linearizer to async (#12357)
Refactor and convert `Linearizer` to async. This makes a `Linearizer`
cancellation bug easier to fix.

Also refactor to use an async context manager, which eliminates an
unlikely footgun where code that doesn't immediately use the context
manager could forget to release the lock.

Signed-off-by: Sean Quah <seanq@element.io>
2022-04-05 15:43:52 +01:00

733 lines
25 KiB
Python

# Copyright 2014-2016 OpenMarket Ltd
# Copyright 2018 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 abc
import collections
import inspect
import itertools
import logging
from contextlib import asynccontextmanager
from typing import (
Any,
AsyncIterator,
Awaitable,
Callable,
Collection,
Dict,
Generic,
Hashable,
Iterable,
List,
Optional,
Set,
Tuple,
TypeVar,
Union,
cast,
overload,
)
import attr
from typing_extensions import AsyncContextManager, Literal
from twisted.internet import defer
from twisted.internet.defer import CancelledError
from twisted.internet.interfaces import IReactorTime
from twisted.python.failure import Failure
from synapse.logging.context import (
PreserveLoggingContext,
make_deferred_yieldable,
run_in_background,
)
from synapse.util import Clock
logger = logging.getLogger(__name__)
_T = TypeVar("_T")
class AbstractObservableDeferred(Generic[_T], metaclass=abc.ABCMeta):
"""Abstract base class defining the consumer interface of ObservableDeferred"""
__slots__ = ()
@abc.abstractmethod
def observe(self) -> "defer.Deferred[_T]":
"""Add a new observer for this ObservableDeferred
This returns a brand new deferred that is resolved when the underlying
deferred is resolved. Interacting with the returned deferred does not
effect the underlying deferred.
Note that the returned Deferred doesn't follow the Synapse logcontext rules -
you will probably want to `make_deferred_yieldable` it.
"""
...
class ObservableDeferred(Generic[_T], AbstractObservableDeferred[_T]):
"""Wraps a deferred object so that we can add observer deferreds. These
observer deferreds do not affect the callback chain of the original
deferred.
If consumeErrors is true errors will be captured from the origin deferred.
Cancelling or otherwise resolving an observer will not affect the original
ObservableDeferred.
NB that it does not attempt to do anything with logcontexts; in general
you should probably make_deferred_yieldable the deferreds
returned by `observe`, and ensure that the original deferred runs its
callbacks in the sentinel logcontext.
"""
__slots__ = ["_deferred", "_observers", "_result"]
_deferred: "defer.Deferred[_T]"
_observers: Union[List["defer.Deferred[_T]"], Tuple[()]]
_result: Union[None, Tuple[Literal[True], _T], Tuple[Literal[False], Failure]]
def __init__(self, deferred: "defer.Deferred[_T]", consumeErrors: bool = False):
object.__setattr__(self, "_deferred", deferred)
object.__setattr__(self, "_result", None)
object.__setattr__(self, "_observers", [])
def callback(r: _T) -> _T:
object.__setattr__(self, "_result", (True, r))
# once we have set _result, no more entries will be added to _observers,
# so it's safe to replace it with the empty tuple.
observers = self._observers
object.__setattr__(self, "_observers", ())
for observer in observers:
try:
observer.callback(r)
except Exception as e:
logger.exception(
"%r threw an exception on .callback(%r), ignoring...",
observer,
r,
exc_info=e,
)
return r
def errback(f: Failure) -> Optional[Failure]:
object.__setattr__(self, "_result", (False, f))
# once we have set _result, no more entries will be added to _observers,
# so it's safe to replace it with the empty tuple.
observers = self._observers
object.__setattr__(self, "_observers", ())
for observer in observers:
# This is a little bit of magic to correctly propagate stack
# traces when we `await` on one of the observer deferreds.
f.value.__failure__ = f # type: ignore[union-attr]
try:
observer.errback(f)
except Exception as e:
logger.exception(
"%r threw an exception on .errback(%r), ignoring...",
observer,
f,
exc_info=e,
)
if consumeErrors:
return None
else:
return f
deferred.addCallbacks(callback, errback)
def observe(self) -> "defer.Deferred[_T]":
"""Observe the underlying deferred.
This returns a brand new deferred that is resolved when the underlying
deferred is resolved. Interacting with the returned deferred does not
effect the underlying deferred.
"""
if not self._result:
assert isinstance(self._observers, list)
d: "defer.Deferred[_T]" = defer.Deferred()
self._observers.append(d)
return d
elif self._result[0]:
return defer.succeed(self._result[1])
else:
return defer.fail(self._result[1])
def observers(self) -> "Collection[defer.Deferred[_T]]":
return self._observers
def has_called(self) -> bool:
return self._result is not None
def has_succeeded(self) -> bool:
return self._result is not None and self._result[0] is True
def get_result(self) -> Union[_T, Failure]:
if self._result is None:
raise ValueError(f"{self!r} has no result yet")
return self._result[1]
def __getattr__(self, name: str) -> Any:
return getattr(self._deferred, name)
def __setattr__(self, name: str, value: Any) -> None:
setattr(self._deferred, name, value)
def __repr__(self) -> str:
return "<ObservableDeferred object at %s, result=%r, _deferred=%r>" % (
id(self),
self._result,
self._deferred,
)
T = TypeVar("T")
async def concurrently_execute(
func: Callable[[T], Any], args: Iterable[T], limit: int
) -> None:
"""Executes the function with each argument concurrently while limiting
the number of concurrent executions.
Args:
func: Function to execute, should return a deferred or coroutine.
args: List of arguments to pass to func, each invocation of func
gets a single argument.
limit: Maximum number of conccurent executions.
Returns:
Deferred: Resolved when all function invocations have finished.
"""
it = iter(args)
async def _concurrently_execute_inner(value: T) -> None:
try:
while True:
await maybe_awaitable(func(value))
value = next(it)
except StopIteration:
pass
# We use `itertools.islice` to handle the case where the number of args is
# less than the limit, avoiding needlessly spawning unnecessary background
# tasks.
await yieldable_gather_results(
_concurrently_execute_inner, (value for value in itertools.islice(it, limit))
)
async def yieldable_gather_results(
func: Callable[..., Awaitable[T]], iter: Iterable, *args: Any, **kwargs: Any
) -> List[T]:
"""Executes the function with each argument concurrently.
Args:
func: Function to execute that returns a Deferred
iter: An iterable that yields items that get passed as the first
argument to the function
*args: Arguments to be passed to each call to func
**kwargs: Keyword arguments to be passed to each call to func
Returns
A list containing the results of the function
"""
try:
return await make_deferred_yieldable(
defer.gatherResults(
[run_in_background(func, item, *args, **kwargs) for item in iter],
consumeErrors=True,
)
)
except defer.FirstError as dfe:
# unwrap the error from defer.gatherResults.
# The raised exception's traceback only includes func() etc if
# the 'await' happens before the exception is thrown - ie if the failure
# happens *asynchronously* - otherwise Twisted throws away the traceback as it
# could be large.
#
# We could maybe reconstruct a fake traceback from Failure.frames. Or maybe
# we could throw Twisted into the fires of Mordor.
# suppress exception chaining, because the FirstError doesn't tell us anything
# very interesting.
assert isinstance(dfe.subFailure.value, BaseException)
raise dfe.subFailure.value from None
T1 = TypeVar("T1")
T2 = TypeVar("T2")
T3 = TypeVar("T3")
@overload
def gather_results(
deferredList: Tuple[()], consumeErrors: bool = ...
) -> "defer.Deferred[Tuple[()]]":
...
@overload
def gather_results(
deferredList: Tuple["defer.Deferred[T1]"],
consumeErrors: bool = ...,
) -> "defer.Deferred[Tuple[T1]]":
...
@overload
def gather_results(
deferredList: Tuple["defer.Deferred[T1]", "defer.Deferred[T2]"],
consumeErrors: bool = ...,
) -> "defer.Deferred[Tuple[T1, T2]]":
...
@overload
def gather_results(
deferredList: Tuple[
"defer.Deferred[T1]", "defer.Deferred[T2]", "defer.Deferred[T3]"
],
consumeErrors: bool = ...,
) -> "defer.Deferred[Tuple[T1, T2, T3]]":
...
def gather_results( # type: ignore[misc]
deferredList: Tuple["defer.Deferred[T1]", ...],
consumeErrors: bool = False,
) -> "defer.Deferred[Tuple[T1, ...]]":
"""Combines a tuple of `Deferred`s into a single `Deferred`.
Wraps `defer.gatherResults` to provide type annotations that support heterogenous
lists of `Deferred`s.
"""
# The `type: ignore[misc]` above suppresses
# "Overloaded function implementation cannot produce return type of signature 1/2/3"
deferred = defer.gatherResults(deferredList, consumeErrors=consumeErrors)
return deferred.addCallback(tuple)
@attr.s(slots=True, auto_attribs=True)
class _LinearizerEntry:
# The number of things executing.
count: int
# Deferreds for the things blocked from executing.
deferreds: collections.OrderedDict
class Linearizer:
"""Limits concurrent access to resources based on a key. Useful to ensure
only a few things happen at a time on a given resource.
Example:
async with limiter.queue("test_key"):
# do some work.
"""
def __init__(
self,
name: Optional[str] = None,
max_count: int = 1,
clock: Optional[Clock] = None,
):
"""
Args:
max_count: The maximum number of concurrent accesses
"""
if name is None:
self.name: Union[str, int] = id(self)
else:
self.name = name
if not clock:
from twisted.internet import reactor
clock = Clock(cast(IReactorTime, reactor))
self._clock = clock
self.max_count = max_count
# key_to_defer is a map from the key to a _LinearizerEntry.
self.key_to_defer: Dict[Hashable, _LinearizerEntry] = {}
def is_queued(self, key: Hashable) -> bool:
"""Checks whether there is a process queued up waiting"""
entry = self.key_to_defer.get(key)
if not entry:
# No entry so nothing is waiting.
return False
# There are waiting deferreds only in the OrderedDict of deferreds is
# non-empty.
return bool(entry.deferreds)
def queue(self, key: Hashable) -> AsyncContextManager[None]:
@asynccontextmanager
async def _ctx_manager() -> AsyncIterator[None]:
entry = await self._acquire_lock(key)
try:
yield
finally:
self._release_lock(key, entry)
return _ctx_manager()
async def _acquire_lock(self, key: Hashable) -> _LinearizerEntry:
"""Acquires a linearizer lock, waiting if necessary.
Returns once we have secured the lock.
"""
entry = self.key_to_defer.setdefault(
key, _LinearizerEntry(0, collections.OrderedDict())
)
if entry.count < self.max_count:
# The number of things executing is less than the maximum.
logger.debug(
"Acquired uncontended linearizer lock %r for key %r", self.name, key
)
entry.count += 1
return entry
# Otherwise, the number of things executing is at the maximum and we have to
# add a deferred to the list of blocked items.
# When one of the things currently executing finishes it will callback
# this item so that it can continue executing.
logger.debug("Waiting to acquire linearizer lock %r for key %r", self.name, key)
new_defer: "defer.Deferred[None]" = make_deferred_yieldable(defer.Deferred())
entry.deferreds[new_defer] = 1
try:
await new_defer
except Exception as e:
logger.info("defer %r got err %r", new_defer, e)
if isinstance(e, CancelledError):
logger.debug(
"Cancelling wait for linearizer lock %r for key %r",
self.name,
key,
)
else:
logger.warning(
"Unexpected exception waiting for linearizer lock %r for key %r",
self.name,
key,
)
# we just have to take ourselves back out of the queue.
del entry.deferreds[new_defer]
raise
logger.debug("Acquired linearizer lock %r for key %r", self.name, key)
entry.count += 1
# if the code holding the lock completes synchronously, then it
# will recursively run the next claimant on the list. That can
# relatively rapidly lead to stack exhaustion. This is essentially
# the same problem as http://twistedmatrix.com/trac/ticket/9304.
#
# In order to break the cycle, we add a cheeky sleep(0) here to
# ensure that we fall back to the reactor between each iteration.
#
# This needs to happen while we hold the lock. We could put it on the
# exit path, but that would slow down the uncontended case.
await self._clock.sleep(0)
return entry
def _release_lock(self, key: Hashable, entry: _LinearizerEntry) -> None:
"""Releases a held linearizer lock."""
logger.debug("Releasing linearizer lock %r for key %r", self.name, key)
# We've finished executing so check if there are any things
# blocked waiting to execute and start one of them
entry.count -= 1
if entry.deferreds:
(next_def, _) = entry.deferreds.popitem(last=False)
# we need to run the next thing in the sentinel context.
with PreserveLoggingContext():
next_def.callback(None)
elif entry.count == 0:
# We were the last thing for this key: remove it from the
# map.
del self.key_to_defer[key]
class ReadWriteLock:
"""An async read write lock.
Example:
async with read_write_lock.read("test_key"):
# do some work
"""
# IMPLEMENTATION NOTES
#
# We track the most recent queued reader and writer deferreds (which get
# resolved when they release the lock).
#
# Read: We know its safe to acquire a read lock when the latest writer has
# been resolved. The new reader is appended to the list of latest readers.
#
# Write: We know its safe to acquire the write lock when both the latest
# writers and readers have been resolved. The new writer replaces the latest
# writer.
def __init__(self) -> None:
# Latest readers queued
self.key_to_current_readers: Dict[str, Set[defer.Deferred]] = {}
# Latest writer queued
self.key_to_current_writer: Dict[str, defer.Deferred] = {}
def read(self, key: str) -> AsyncContextManager:
@asynccontextmanager
async def _ctx_manager() -> AsyncIterator[None]:
new_defer: "defer.Deferred[None]" = defer.Deferred()
curr_readers = self.key_to_current_readers.setdefault(key, set())
curr_writer = self.key_to_current_writer.get(key, None)
curr_readers.add(new_defer)
try:
# We wait for the latest writer to finish writing. We can safely ignore
# any existing readers... as they're readers.
# May raise a `CancelledError` if the `Deferred` wrapping us is
# cancelled. The `Deferred` we are waiting on must not be cancelled,
# since we do not own it.
if curr_writer:
await make_deferred_yieldable(stop_cancellation(curr_writer))
yield
finally:
with PreserveLoggingContext():
new_defer.callback(None)
self.key_to_current_readers.get(key, set()).discard(new_defer)
return _ctx_manager()
def write(self, key: str) -> AsyncContextManager:
@asynccontextmanager
async def _ctx_manager() -> AsyncIterator[None]:
new_defer: "defer.Deferred[None]" = defer.Deferred()
curr_readers = self.key_to_current_readers.get(key, set())
curr_writer = self.key_to_current_writer.get(key, None)
# We wait on all latest readers and writer.
to_wait_on = list(curr_readers)
if curr_writer:
to_wait_on.append(curr_writer)
# We can clear the list of current readers since `new_defer` waits
# for them to finish.
curr_readers.clear()
self.key_to_current_writer[key] = new_defer
to_wait_on_defer = defer.gatherResults(to_wait_on)
try:
# Wait for all current readers and the latest writer to finish.
# May raise a `CancelledError` immediately after the wait if the
# `Deferred` wrapping us is cancelled. We must only release the lock
# once we have acquired it, hence the use of `delay_cancellation`
# rather than `stop_cancellation`.
await make_deferred_yieldable(delay_cancellation(to_wait_on_defer))
yield
finally:
# Release the lock.
with PreserveLoggingContext():
new_defer.callback(None)
# `self.key_to_current_writer[key]` may be missing if there was another
# writer waiting for us and it completed entirely within the
# `new_defer.callback()` call above.
if self.key_to_current_writer.get(key) == new_defer:
self.key_to_current_writer.pop(key)
return _ctx_manager()
R = TypeVar("R")
def timeout_deferred(
deferred: "defer.Deferred[_T]", timeout: float, reactor: IReactorTime
) -> "defer.Deferred[_T]":
"""The in built twisted `Deferred.addTimeout` fails to time out deferreds
that have a canceller that throws exceptions. This method creates a new
deferred that wraps and times out the given deferred, correctly handling
the case where the given deferred's canceller throws.
(See https://twistedmatrix.com/trac/ticket/9534)
NOTE: Unlike `Deferred.addTimeout`, this function returns a new deferred.
NOTE: the TimeoutError raised by the resultant deferred is
twisted.internet.defer.TimeoutError, which is *different* to the built-in
TimeoutError, as well as various other TimeoutErrors you might have imported.
Args:
deferred: The Deferred to potentially timeout.
timeout: Timeout in seconds
reactor: The twisted reactor to use
Returns:
A new Deferred, which will errback with defer.TimeoutError on timeout.
"""
new_d: "defer.Deferred[_T]" = defer.Deferred()
timed_out = [False]
def time_it_out() -> None:
timed_out[0] = True
try:
deferred.cancel()
except Exception: # if we throw any exception it'll break time outs
logger.exception("Canceller failed during timeout")
# the cancel() call should have set off a chain of errbacks which
# will have errbacked new_d, but in case it hasn't, errback it now.
if not new_d.called:
new_d.errback(defer.TimeoutError("Timed out after %gs" % (timeout,)))
delayed_call = reactor.callLater(timeout, time_it_out)
def convert_cancelled(value: Failure) -> Failure:
# if the original deferred was cancelled, and our timeout has fired, then
# the reason it was cancelled was due to our timeout. Turn the CancelledError
# into a TimeoutError.
if timed_out[0] and value.check(CancelledError):
raise defer.TimeoutError("Timed out after %gs" % (timeout,))
return value
deferred.addErrback(convert_cancelled)
def cancel_timeout(result: _T) -> _T:
# stop the pending call to cancel the deferred if it's been fired
if delayed_call.active():
delayed_call.cancel()
return result
deferred.addBoth(cancel_timeout)
def success_cb(val: _T) -> None:
if not new_d.called:
new_d.callback(val)
def failure_cb(val: Failure) -> None:
if not new_d.called:
new_d.errback(val)
deferred.addCallbacks(success_cb, failure_cb)
return new_d
# This class can't be generic because it uses slots with attrs.
# See: https://github.com/python-attrs/attrs/issues/313
@attr.s(slots=True, frozen=True, auto_attribs=True)
class DoneAwaitable: # should be: Generic[R]
"""Simple awaitable that returns the provided value."""
value: Any # should be: R
def __await__(self) -> Any:
return self
def __iter__(self) -> "DoneAwaitable":
return self
def __next__(self) -> None:
raise StopIteration(self.value)
def maybe_awaitable(value: Union[Awaitable[R], R]) -> Awaitable[R]:
"""Convert a value to an awaitable if not already an awaitable."""
if inspect.isawaitable(value):
assert isinstance(value, Awaitable)
return value
return DoneAwaitable(value)
def stop_cancellation(deferred: "defer.Deferred[T]") -> "defer.Deferred[T]":
"""Prevent a `Deferred` from being cancelled by wrapping it in another `Deferred`.
Args:
deferred: The `Deferred` to protect against cancellation. Must not follow the
Synapse logcontext rules.
Returns:
A new `Deferred`, which will contain the result of the original `Deferred`.
The new `Deferred` will not propagate cancellation through to the original.
When cancelled, the new `Deferred` will fail with a `CancelledError`.
The new `Deferred` will not follow the Synapse logcontext rules and should be
wrapped with `make_deferred_yieldable`.
"""
new_deferred: "defer.Deferred[T]" = defer.Deferred()
deferred.chainDeferred(new_deferred)
return new_deferred
def delay_cancellation(deferred: "defer.Deferred[T]") -> "defer.Deferred[T]":
"""Delay cancellation of a `Deferred` until it resolves.
Has the same effect as `stop_cancellation`, but the returned `Deferred` will not
resolve with a `CancelledError` until the original `Deferred` resolves.
Args:
deferred: The `Deferred` to protect against cancellation. May optionally follow
the Synapse logcontext rules.
Returns:
A new `Deferred`, which will contain the result of the original `Deferred`.
The new `Deferred` will not propagate cancellation through to the original.
When cancelled, the new `Deferred` will wait until the original `Deferred`
resolves before failing with a `CancelledError`.
The new `Deferred` will follow the Synapse logcontext rules if `deferred`
follows the Synapse logcontext rules. Otherwise the new `Deferred` should be
wrapped with `make_deferred_yieldable`.
"""
def handle_cancel(new_deferred: "defer.Deferred[T]") -> None:
# before the new deferred is cancelled, we `pause` it to stop the cancellation
# propagating. we then `unpause` it once the wrapped deferred completes, to
# propagate the exception.
new_deferred.pause()
new_deferred.errback(Failure(CancelledError()))
deferred.addBoth(lambda _: new_deferred.unpause())
new_deferred: "defer.Deferred[T]" = defer.Deferred(handle_cancel)
deferred.chainDeferred(new_deferred)
return new_deferred