dandri/synapse
1
0
Fork 0
mirror of https://github.com/element-hq/synapse.git synced 2026-04-25 19:42:23 +00:00
Code Issues Packages Projects Releases Wiki Activity
Files
be52e60bf1fa7a9be23f625fbb00b71242674d03
synapse/tests/storage/databases/main/test_lock.py
Matthew Hodgson ac26cfac83 Let me give you a clear summary of where this project stands: 
---
  Summary: Synapse Twisted → asyncio Migration

  What's been built (Phases 0-7) — 10 new files, ~3200 lines, 96 tests

  Every Twisted component has an asyncio-native replacement ready:

  ┌──────────────────┬───────────────────────────────────────────────┬─────────────────────────────────────────────────────────┬───────────────────────────────────────────────┐
  │    Component     │               Twisted Original                │                   asyncio Replacement                   │                     File                      │
  ├──────────────────┼───────────────────────────────────────────────┼─────────────────────────────────────────────────────────┼───────────────────────────────────────────────┤
  │ LoggingContext   │ threading.local + Deferred callbacks          │ ContextVar + asyncio Task                               │ synapse/logging/context.py (native functions) │
  ├──────────────────┼───────────────────────────────────────────────┼─────────────────────────────────────────────────────────┼───────────────────────────────────────────────┤
  │ Async primitives │ ObservableDeferred, Linearizer, ReadWriteLock │ ObservableFuture, NativeLinearizer, NativeReadWriteLock │ synapse/util/async_helpers.py                 │
  ├──────────────────┼───────────────────────────────────────────────┼─────────────────────────────────────────────────────────┼───────────────────────────────────────────────┤
  │ Cache            │ DeferredCache                                 │ FutureCache                                             │ synapse/util/caches/future_cache.py           │
  ├──────────────────┼───────────────────────────────────────────────┼─────────────────────────────────────────────────────────┼───────────────────────────────────────────────┤
  │ Clock            │ Clock (Twisted reactor)                       │ NativeClock (asyncio)                                   │ synapse/util/clock.py                         │
  ├──────────────────┼───────────────────────────────────────────────┼─────────────────────────────────────────────────────────┼───────────────────────────────────────────────┤
  │ Database         │ adbapi.ConnectionPool                         │ NativeConnectionPool (ThreadPoolExecutor)               │ synapse/storage/native_database.py            │
  ├──────────────────┼───────────────────────────────────────────────┼─────────────────────────────────────────────────────────┼───────────────────────────────────────────────┤
  │ HTTP Client      │ treq + Twisted Agent                          │ aiohttp.ClientSession                                   │ synapse/http/native_client.py                 │
  ├──────────────────┼───────────────────────────────────────────────┼─────────────────────────────────────────────────────────┼───────────────────────────────────────────────┤
  │ HTTP Server      │ JsonResource + Twisted Site                   │ NativeJsonResource + aiohttp.web                        │ synapse/http/native_server.py                 │
  ├──────────────────┼───────────────────────────────────────────────┼─────────────────────────────────────────────────────────┼───────────────────────────────────────────────┤
  │ Replication      │ LineOnlyReceiver (Twisted Protocol)           │ asyncio.StreamReader/Writer                             │ synapse/replication/tcp/native_protocol.py    │
  ├──────────────────┼───────────────────────────────────────────────┼─────────────────────────────────────────────────────────┼───────────────────────────────────────────────┤
  │ Event loop       │ ISynapseReactor                               │ ISynapseEventLoop                                       │ synapse/types/__init__.py                     │
  ├──────────────────┼───────────────────────────────────────────────┼─────────────────────────────────────────────────────────┼───────────────────────────────────────────────┤
  │ Test helper      │ —                                             │ FakeAsyncioLoop                                         │ tests/async_helpers.py                        │
  └──────────────────┴───────────────────────────────────────────────┴─────────────────────────────────────────────────────────┴───────────────────────────────────────────────┘

  What's been wired in safely — 224 files changed, 0 regressions

  - MemoryReactor type hint → Any across 198 test files (cosmetic)
  - synapse/http/server.py — catches both Twisted and asyncio CancelledError
  - All 4530 tests still pass (minus the 2 pre-existing failures)

  What remains for the flag day

  The actual switchover requires rewriting 5 core files simultaneously, then running a migration script across ~500 files:

  1. tests/unittest.py + tests/server.py — switch from twisted.trial.TestCase to unittest.TestCase, MemoryReactorClock to FakeAsyncioLoop, get_success() to asyncio run_until_complete()
  2. synapse/logging/context.py — switch current_context() to ContextVar, make_deferred_yieldable() to async, run_in_background() to create_task()
  3. synapse/util/async_helpers.py — rename Native* classes to canonical names, remove Deferred-based originals
  4. Migration script — update all CancelledError, defer.*, Deferred imports across ~500 files
  5. pyproject.toml — remove Twisted dependency

  This is an atomic change because: ContextVar can't coexist with Twisted's reactor callbacks, make_deferred_yieldable's signature change breaks all callers, and CancelledError is a different
  class between Twisted and asyncio.
2026-03-21 16:17:04 +00:00

500 lines
16 KiB
Python
Raw Blame History

#
# This file is licensed under the Affero General Public License (AGPL) version 3.
#
# Copyright 2021 The Matrix.org Foundation C.I.C.
# Copyright (C) 2023 New Vector, Ltd
#
# This program is free software: you can redistribute it and/or modify
# it under the terms of the GNU Affero General Public License as
# published by the Free Software Foundation, either version 3 of the
# License, or (at your option) any later version.
#
# See the GNU Affero General Public License for more details:
# <https://www.gnu.org/licenses/agpl-3.0.html>.
#
# Originally licensed under the Apache License, Version 2.0:
# <http://www.apache.org/licenses/LICENSE-2.0>.
#
# [This file includes modifications made by New Vector Limited]
#
#
from twisted.internet import defer, reactor
from twisted.internet.base import ReactorBase
from twisted.internet.defer import Deferred
from typing import Any as MemoryReactor # was: MemoryReactor from Twisted
from synapse.server import HomeServer
from synapse.storage.databases.main.lock import _LOCK_TIMEOUT_MS, _RENEWAL_INTERVAL
from synapse.util.clock import Clock
from tests import unittest
class LockTestCase(unittest.HomeserverTestCase):
def prepare(self, reactor: MemoryReactor, clock: Clock, hs: HomeServer) -> None:
self.store = hs.get_datastores().main
def test_acquire_contention(self) -> None:
# Track the number of tasks holding the lock.
# Should be at most 1.
in_lock = 0
max_in_lock = 0
release_lock: "Deferred[None]" = Deferred()
async def task() -> None:
nonlocal in_lock
nonlocal max_in_lock
lock = await self.store.try_acquire_lock("name", "key")
if not lock:
return
async with lock:
in_lock += 1
max_in_lock = max(max_in_lock, in_lock)
# Block to allow other tasks to attempt to take the lock.
await release_lock
in_lock -= 1
# Start 3 tasks.
task1 = defer.ensureDeferred(task())
task2 = defer.ensureDeferred(task())
task3 = defer.ensureDeferred(task())
# Give the reactor a kick so that the database transaction returns.
self.pump()
release_lock.callback(None)
# Run the tasks to completion.
self.get_success(task1)
self.pump()
self.get_success(task2)
self.pump()
self.get_success(task3)
# At most one task should have held the lock at a time.
self.assertEqual(max_in_lock, 1)
def test_simple_lock(self) -> None:
"""Test that we can take out a lock and that while we hold it nobody
else can take it out.
"""
# First to acquire this lock, so it should complete
lock = self.get_success(self.store.try_acquire_lock("name", "key"))
assert lock is not None
# Enter the context manager
self.get_success(lock.__aenter__())
# Attempting to acquire the lock again fails.
lock2 = self.get_success(self.store.try_acquire_lock("name", "key"))
self.assertIsNone(lock2)
# Calling `is_still_valid` reports true.
self.assertTrue(self.get_success(lock.is_still_valid()))
# Drop the lock
self.get_success(lock.__aexit__(None, None, None))
# We can now acquire the lock again.
lock3 = self.get_success(self.store.try_acquire_lock("name", "key"))
assert lock3 is not None
self.get_success(lock3.__aenter__())
self.get_success(lock3.__aexit__(None, None, None))
def test_maintain_lock(self) -> None:
"""Test that we don't time out locks while they're still active"""
lock = self.get_success(self.store.try_acquire_lock("name", "key"))
assert lock is not None
self.get_success(lock.__aenter__())
# Wait for ages with the lock, we should not be able to get the lock.
self.reactor.advance(5 * _LOCK_TIMEOUT_MS / 1000)
lock2 = self.get_success(self.store.try_acquire_lock("name", "key"))
self.assertIsNone(lock2)
self.get_success(lock.__aexit__(None, None, None))
def test_timeout_lock(self) -> None:
"""Test that we time out locks if they're not updated for ages"""
lock = self.get_success(self.store.try_acquire_lock("name", "key"))
assert lock is not None
self.get_success(lock.__aenter__())
# We simulate the process getting stuck by cancelling the looping call
# that keeps the lock active.
assert lock._looping_call
lock._looping_call.stop()
# Wait for the lock to timeout.
self.reactor.advance(2 * _LOCK_TIMEOUT_MS / 1000)
lock2 = self.get_success(self.store.try_acquire_lock("name", "key"))
self.assertIsNotNone(lock2)
self.assertFalse(self.get_success(lock.is_still_valid()))
def test_drop(self) -> None:
"""Test that dropping the context manager means we stop renewing the lock"""
lock = self.get_success(self.store.try_acquire_lock("name", "key"))
self.assertIsNotNone(lock)
del lock
# Wait for the lock to timeout.
self.reactor.advance(2 * _LOCK_TIMEOUT_MS / 1000)
lock2 = self.get_success(self.store.try_acquire_lock("name", "key"))
self.assertIsNotNone(lock2)
def test_shutdown(self) -> None:
"""Test that shutting down Synapse releases the locks"""
# Acquire two locks
lock = self.get_success(self.store.try_acquire_lock("name", "key1"))
self.assertIsNotNone(lock)
lock2 = self.get_success(self.store.try_acquire_lock("name", "key2"))
self.assertIsNotNone(lock2)
# Now call the shutdown code
self.get_success(self.store._on_shutdown())
self.assertEqual(self.store._live_lock_tokens, {})
class ReadWriteLockTestCase(unittest.HomeserverTestCase):
"""Test the read/write lock implementation."""
def prepare(self, reactor: MemoryReactor, clock: Clock, hs: HomeServer) -> None:
self.store = hs.get_datastores().main
def test_acquire_write_contention(self) -> None:
"""Test that we can only acquire one write lock at a time"""
# Track the number of tasks holding the lock.
# Should be at most 1.
in_lock = 0
max_in_lock = 0
release_lock: "Deferred[None]" = Deferred()
async def task() -> None:
nonlocal in_lock
nonlocal max_in_lock
lock = await self.store.try_acquire_read_write_lock(
"name", "key", write=True
)
if not lock:
return
async with lock:
in_lock += 1
max_in_lock = max(max_in_lock, in_lock)
# Block to allow other tasks to attempt to take the lock.
await release_lock
in_lock -= 1
# Start 3 tasks.
task1 = defer.ensureDeferred(task())
task2 = defer.ensureDeferred(task())
task3 = defer.ensureDeferred(task())
# Give the reactor a kick so that the database transaction returns.
self.pump()
release_lock.callback(None)
# Run the tasks to completion.
assert isinstance(reactor, ReactorBase)
self.get_success(task1)
self.pump()
self.get_success(task2)
self.pump()
self.get_success(task3)
# At most one task should have held the lock at a time.
self.assertEqual(max_in_lock, 1)
def test_acquire_multiple_reads(self) -> None:
"""Test that we can acquire multiple read locks at a time"""
# Track the number of tasks holding the lock.
in_lock = 0
max_in_lock = 0
release_lock: "Deferred[None]" = Deferred()
async def task() -> None:
nonlocal in_lock
nonlocal max_in_lock
lock = await self.store.try_acquire_read_write_lock(
"name", "key", write=False
)
if not lock:
return
async with lock:
in_lock += 1
max_in_lock = max(max_in_lock, in_lock)
# Block to allow other tasks to attempt to take the lock.
await release_lock
in_lock -= 1
# Start 3 tasks.
task1 = defer.ensureDeferred(task())
task2 = defer.ensureDeferred(task())
task3 = defer.ensureDeferred(task())
# Give the reactor a kick so that the database transaction returns.
self.pump()
release_lock.callback(None)
# Run the tasks to completion.
assert isinstance(reactor, ReactorBase)
self.get_success(task1)
self.pump()
self.get_success(task2)
self.pump()
self.get_success(task3)
# At most one task should have held the lock at a time.
self.assertEqual(max_in_lock, 3)
def test_write_lock_acquired(self) -> None:
"""Test that we can take out a write lock and that while we hold it
nobody else can take it out.
"""
# First to acquire this lock, so it should complete
lock = self.get_success(
self.store.try_acquire_read_write_lock("name", "key", write=True)
)
assert lock is not None
# Enter the context manager
self.get_success(lock.__aenter__())
# Attempting to acquire the lock again fails, as both read and write.
lock2 = self.get_success(
self.store.try_acquire_read_write_lock("name", "key", write=True)
)
self.assertIsNone(lock2)
lock3 = self.get_success(
self.store.try_acquire_read_write_lock("name", "key", write=False)
)
self.assertIsNone(lock3)
# Calling `is_still_valid` reports true.
self.assertTrue(self.get_success(lock.is_still_valid()))
# Drop the lock
self.get_success(lock.__aexit__(None, None, None))
# We can now acquire the lock again.
lock4 = self.get_success(
self.store.try_acquire_read_write_lock("name", "key", write=True)
)
assert lock4 is not None
self.get_success(lock4.__aenter__())
self.get_success(lock4.__aexit__(None, None, None))
def test_read_lock_acquired(self) -> None:
"""Test that we can take out a read lock and that while we hold it
only other reads can use it.
"""
# First to acquire this lock, so it should complete
lock = self.get_success(
self.store.try_acquire_read_write_lock("name", "key", write=False)
)
assert lock is not None
# Enter the context manager
self.get_success(lock.__aenter__())
# Attempting to acquire the write lock fails
lock2 = self.get_success(
self.store.try_acquire_read_write_lock("name", "key", write=True)
)
self.assertIsNone(lock2)
# Attempting to acquire a read lock succeeds
lock3 = self.get_success(
self.store.try_acquire_read_write_lock("name", "key", write=False)
)
assert lock3 is not None
self.get_success(lock3.__aenter__())
# Calling `is_still_valid` reports true.
self.assertTrue(self.get_success(lock.is_still_valid()))
# Drop the first lock
self.get_success(lock.__aexit__(None, None, None))
# Attempting to acquire the write lock still fails, as lock3 is still
# active.
lock4 = self.get_success(
self.store.try_acquire_read_write_lock("name", "key", write=True)
)
self.assertIsNone(lock4)
# Drop the still open third lock
self.get_success(lock3.__aexit__(None, None, None))
# We can now acquire the lock again.
lock5 = self.get_success(
self.store.try_acquire_read_write_lock("name", "key", write=True)
)
assert lock5 is not None
self.get_success(lock5.__aenter__())
self.get_success(lock5.__aexit__(None, None, None))
def test_maintain_lock(self) -> None:
"""Test that we don't time out locks while they're still active (lock is
renewed in the background if the process is still alive)"""
lock = self.get_success(
self.store.try_acquire_read_write_lock("name", "key", write=True)
)
assert lock is not None
self.get_success(lock.__aenter__())
# Wait for ages with the lock, we should not be able to get the lock.
for _ in range(10):
self.reactor.advance((_RENEWAL_INTERVAL.as_secs()))
lock2 = self.get_success(
self.store.try_acquire_read_write_lock("name", "key", write=True)
)
self.assertIsNone(lock2)
self.get_success(lock.__aexit__(None, None, None))
def test_timeout_lock(self) -> None:
"""Test that we time out locks if they're not updated for ages"""
lock = self.get_success(
self.store.try_acquire_read_write_lock("name", "key", write=True)
)
assert lock is not None
self.get_success(lock.__aenter__())
# We simulate the process getting stuck by cancelling the looping call
# that keeps the lock active.
assert lock._looping_call
lock._looping_call.stop()
# Wait for the lock to timeout.
self.reactor.advance(2 * _LOCK_TIMEOUT_MS / 1000)
lock2 = self.get_success(
self.store.try_acquire_read_write_lock("name", "key", write=True)
)
self.assertIsNotNone(lock2)
self.assertFalse(self.get_success(lock.is_still_valid()))
def test_drop(self) -> None:
"""Test that dropping the context manager means we stop renewing the lock"""
lock = self.get_success(
self.store.try_acquire_read_write_lock("name", "key", write=True)
)
self.assertIsNotNone(lock)
del lock
# Wait for the lock to timeout.
self.reactor.advance(2 * _LOCK_TIMEOUT_MS / 1000)
lock2 = self.get_success(
self.store.try_acquire_read_write_lock("name", "key", write=True)
)
self.assertIsNotNone(lock2)
def test_shutdown(self) -> None:
"""Test that shutting down Synapse releases the locks"""
# Acquire two locks
lock = self.get_success(
self.store.try_acquire_read_write_lock("name", "key", write=True)
)
self.assertIsNotNone(lock)
lock2 = self.get_success(
self.store.try_acquire_read_write_lock("name", "key2", write=True)
)
self.assertIsNotNone(lock2)
# Now call the shutdown code
self.get_success(self.store._on_shutdown())
self.assertEqual(self.store._live_read_write_lock_tokens, {})
def test_acquire_multiple_locks(self) -> None:
"""Tests that acquiring multiple locks at once works."""
# Take out multiple locks and ensure that we can't get those locks out
# again.
lock = self.get_success(
self.store.try_acquire_multi_read_write_lock(
[("name1", "key1"), ("name2", "key2")], write=True
)
)
self.assertIsNotNone(lock)
assert lock is not None
self.get_success(lock.__aenter__())
lock2 = self.get_success(
self.store.try_acquire_read_write_lock("name1", "key1", write=True)
)
self.assertIsNone(lock2)
lock3 = self.get_success(
self.store.try_acquire_read_write_lock("name2", "key2", write=False)
)
self.assertIsNone(lock3)
# Overlapping locks attempts will fail, and won't lock any locks.
lock4 = self.get_success(
self.store.try_acquire_multi_read_write_lock(
[("name1", "key1"), ("name3", "key3")], write=True
)
)
self.assertIsNone(lock4)
lock5 = self.get_success(
self.store.try_acquire_read_write_lock("name3", "key3", write=True)
)
self.assertIsNotNone(lock5)
assert lock5 is not None
self.get_success(lock5.__aenter__())
self.get_success(lock5.__aexit__(None, None, None))
# Once we release the lock we can take out the locks again.
self.get_success(lock.__aexit__(None, None, None))
lock6 = self.get_success(
self.store.try_acquire_read_write_lock("name1", "key1", write=True)
)
self.assertIsNotNone(lock6)
assert lock6 is not None
self.get_success(lock6.__aenter__())
self.get_success(lock6.__aexit__(None, None, None))
Reference in New Issue View Git Blame Copy Permalink