dandri/synapse
1
0 Fork 0
mirror of https://github.com/element-hq/synapse.git synced 2026-11-04 17:47:28 +00:00
Code Issues Packages Projects Releases Wiki Activity
Files
ef18e9ebc784ea0adf8a4cd5fc054ffd7f26d8bd
synapse/tests/util/test_lrucache.py
T
Erik JohnstonandClaude Opus 5.5 47367a14a2 Make the Rust event cache EventsWorkerStore's event cache 
`EventsWorkerStore._get_event_cache` is now the Rust `EventCache` that the
sync handler and state resolution read, so each event is held once.

The cache keeps Rust `CachedEvent` values. Python gets them through
`RustEventCache`, a `RustCache` subclass with the `AsyncLruCache` names the
rest of Synapse uses, and each hit wraps the cached event in a new `Event`
object. The wrapper shares the cached event's internal metadata, unsigned
data and signatures, so a change Python makes, such as adding
`prev_content` or flipping `outlier`, lands in the one copy the Rust
readers see. Python's `_event_ref` weak map is checked before the cache,
so while anything holds an event, a lookup returns that same object, as
before. A hit there marks the cache entry as read, and puts the event back
if it was evicted.

- The cache is sized from `caches.event_cache_size` and named
  `*getEvent*`, so its metrics and `per_cache_factors` keep their names.
  The Rust loader now counts its misses too.
- Invalidating an event drops its chain position as well, and a room
  purge drops the room's events through the room index, plus the chain
  cover and state group caches, as `RustStore.invalidate_event` and
  `clear_event_cache` did. Those two calls are gone.
- `EventsWorkerStore` and `RustWorkerStore` run their `__init__` bodies
  in a different order in `DataStore` and `GenericWorkerStore`, as they
  sit differently in each MRO. So whichever runs first builds the
  `RustStore`. A datastore without `RustWorkerStore`, such as the port
  script's, gets a standalone `RustEventCache`.
- The external cache stubs (`get_external`, `set_external`,
  `_invalidate_async_get_event_cache` and its room variant) are dropped.
  They were documented no-ops.
- The sync shadow diff still ignores `prev_content` on state events.
  Python keeps the objects it holds after the cache evicts them, so a Rust
  reader can still load a second copy without it. Trial's cache of one
  event makes that routine.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VpBQxvqZTUvHCMm3u6CR7J
2026-10-15 12:14:54 +00:00

636 lines
21 KiB
Python
Raw Blame History

#
# This file is licensed under the Affero General Public License (AGPL) version 3.
#
# Copyright 2015, 2016 OpenMarket Ltd
# 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 unittest.mock import Mock, patch
from synapse.api.room_versions import RoomVersions
from synapse.events import make_event_from_dict
from synapse.metrics.jemalloc import JemallocStats
from synapse.rest.admin.experimental_features import ExperimentalFeature
from synapse.synapse_rust.storage import RustEventCache
from synapse.types import JsonDict
from synapse.util.caches.lrucache import LruCache, setup_expire_lru_cache_entries
from synapse.util.caches.treecache import TreeCache
from tests import unittest
from tests.server import get_clock
from tests.unittest import override_config
class LruCacheTestCase(unittest.HomeserverTestCase):
def setUp(self) -> None:
super().setUp()
_, self.clock = get_clock()
def test_get_set(self) -> None:
cache: LruCache[str, str] = LruCache(
max_size=1, clock=self.clock, server_name="test_server"
)
cache["key"] = "value"
self.assertEqual(cache.get("key"), "value")
self.assertEqual(cache["key"], "value")
def test_eviction(self) -> None:
cache: LruCache[int, int] = LruCache(
max_size=2, clock=self.clock, server_name="test_server"
)
cache[1] = 1
cache[2] = 2
self.assertEqual(cache.get(1), 1)
self.assertEqual(cache.get(2), 2)
cache[3] = 3
self.assertEqual(cache.get(1), None)
self.assertEqual(cache.get(2), 2)
self.assertEqual(cache.get(3), 3)
def test_setdefault(self) -> None:
cache: LruCache[str, int] = LruCache(
max_size=1, clock=self.clock, server_name="test_server"
)
self.assertEqual(cache.setdefault("key", 1), 1)
self.assertEqual(cache.get("key"), 1)
self.assertEqual(cache.setdefault("key", 2), 1)
self.assertEqual(cache.get("key"), 1)
cache["key"] = 2 # Make sure overriding works.
self.assertEqual(cache.get("key"), 2)
def test_pop(self) -> None:
cache: LruCache[str, int] = LruCache(
max_size=1, clock=self.clock, server_name="test_server"
)
cache["key"] = 1
self.assertEqual(cache.pop("key"), 1)
self.assertEqual(cache.pop("key"), None)
def test_del_multi(self) -> None:
# The type here isn't quite correct as they don't handle TreeCache well.
cache: LruCache[tuple[str, str], str] = LruCache(
max_size=4,
clock=self.clock,
cache_type=TreeCache,
server_name="test_server",
)
cache[("animal", "cat")] = "mew"
cache[("animal", "dog")] = "woof"
cache[("vehicles", "car")] = "vroom"
cache[("vehicles", "train")] = "chuff"
self.assertEqual(len(cache), 4)
self.assertEqual(cache.get(("animal", "cat")), "mew")
self.assertEqual(cache.get(("vehicles", "car")), "vroom")
cache.del_multi(("animal",)) # type: ignore[arg-type]
self.assertEqual(len(cache), 2)
self.assertEqual(cache.get(("animal", "cat")), None)
self.assertEqual(cache.get(("animal", "dog")), None)
self.assertEqual(cache.get(("vehicles", "car")), "vroom")
self.assertEqual(cache.get(("vehicles", "train")), "chuff")
# Man from del_multi say "Yes".
def test_clear(self) -> None:
cache: LruCache[str, int] = LruCache(
max_size=1, clock=self.clock, server_name="test_server"
)
cache["key"] = 1
cache.clear()
self.assertEqual(len(cache), 0)
@override_config({"caches": {"per_cache_factors": {"mycache": 10}}})
def test_special_size(self) -> None:
cache: LruCache = LruCache(
max_size=10,
clock=self.clock,
server_name="test_server",
cache_name="mycache",
)
self.assertEqual(cache.max_size, 100)
class LruCacheCallbacksTestCase(unittest.HomeserverTestCase):
def test_get(self) -> None:
m = Mock()
cache: LruCache[str, str] = LruCache(
max_size=1, clock=self.clock, server_name="test_server"
)
cache.set("key", "value")
self.assertFalse(m.called)
cache.get("key", callbacks=[m])
self.assertFalse(m.called)
cache.get("key", "value")
self.assertFalse(m.called)
cache.set("key", "value2")
self.assertEqual(m.call_count, 1)
cache.set("key", "value")
self.assertEqual(m.call_count, 1)
def test_multi_get(self) -> None:
m = Mock()
cache: LruCache[str, str] = LruCache(
max_size=1, clock=self.clock, server_name="test_server"
)
cache.set("key", "value")
self.assertFalse(m.called)
cache.get("key", callbacks=[m])
self.assertFalse(m.called)
cache.get("key", callbacks=[m])
self.assertFalse(m.called)
cache.set("key", "value2")
self.assertEqual(m.call_count, 1)
cache.set("key", "value")
self.assertEqual(m.call_count, 1)
def test_set(self) -> None:
m = Mock()
cache: LruCache[str, str] = LruCache(
max_size=1, clock=self.clock, server_name="test_server"
)
cache.set("key", "value", callbacks=[m])
self.assertFalse(m.called)
cache.set("key", "value")
self.assertFalse(m.called)
cache.set("key", "value2")
self.assertEqual(m.call_count, 1)
cache.set("key", "value")
self.assertEqual(m.call_count, 1)
def test_pop(self) -> None:
m = Mock()
cache: LruCache[str, str] = LruCache(
max_size=1, clock=self.clock, server_name="test_server"
)
cache.set("key", "value", callbacks=[m])
self.assertFalse(m.called)
cache.pop("key")
self.assertEqual(m.call_count, 1)
cache.set("key", "value")
self.assertEqual(m.call_count, 1)
cache.pop("key")
self.assertEqual(m.call_count, 1)
def test_del_multi(self) -> None:
m1 = Mock()
m2 = Mock()
m3 = Mock()
m4 = Mock()
# The type here isn't quite correct as they don't handle TreeCache well.
cache: LruCache[tuple[str, str], str] = LruCache(
max_size=4,
clock=self.clock,
cache_type=TreeCache,
server_name="test_server",
)
cache.set(("a", "1"), "value", callbacks=[m1])
cache.set(("a", "2"), "value", callbacks=[m2])
cache.set(("b", "1"), "value", callbacks=[m3])
cache.set(("b", "2"), "value", callbacks=[m4])
self.assertEqual(m1.call_count, 0)
self.assertEqual(m2.call_count, 0)
self.assertEqual(m3.call_count, 0)
self.assertEqual(m4.call_count, 0)
cache.del_multi(("a",)) # type: ignore[arg-type]
self.assertEqual(m1.call_count, 1)
self.assertEqual(m2.call_count, 1)
self.assertEqual(m3.call_count, 0)
self.assertEqual(m4.call_count, 0)
def test_clear(self) -> None:
m1 = Mock()
m2 = Mock()
cache: LruCache[str, str] = LruCache(
max_size=5, clock=self.clock, server_name="test_server"
)
cache.set("key1", "value", callbacks=[m1])
cache.set("key2", "value", callbacks=[m2])
self.assertEqual(m1.call_count, 0)
self.assertEqual(m2.call_count, 0)
cache.clear()
self.assertEqual(m1.call_count, 1)
self.assertEqual(m2.call_count, 1)
def test_eviction(self) -> None:
m1 = Mock(name="m1")
m2 = Mock(name="m2")
m3 = Mock(name="m3")
cache: LruCache[str, str] = LruCache(
max_size=2, clock=self.clock, server_name="test_server"
)
cache.set("key1", "value", callbacks=[m1])
cache.set("key2", "value", callbacks=[m2])
self.assertEqual(m1.call_count, 0)
self.assertEqual(m2.call_count, 0)
self.assertEqual(m3.call_count, 0)
cache.set("key3", "value", callbacks=[m3])
self.assertEqual(m1.call_count, 1)
self.assertEqual(m2.call_count, 0)
self.assertEqual(m3.call_count, 0)
cache.set("key3", "value")
self.assertEqual(m1.call_count, 1)
self.assertEqual(m2.call_count, 0)
self.assertEqual(m3.call_count, 0)
cache.get("key2")
self.assertEqual(m1.call_count, 1)
self.assertEqual(m2.call_count, 0)
self.assertEqual(m3.call_count, 0)
cache.set("key1", "value", callbacks=[m1])
self.assertEqual(m1.call_count, 1)
self.assertEqual(m2.call_count, 0)
self.assertEqual(m3.call_count, 1)
class LruCacheSizedTestCase(unittest.HomeserverTestCase):
def test_evict(self) -> None:
cache: LruCache[str, list[int]] = LruCache(
max_size=5, clock=self.clock, size_callback=len, server_name="test_server"
)
cache["key1"] = [0]
cache["key2"] = [1, 2]
cache["key3"] = [3]
cache["key4"] = [4]
self.assertEqual(cache["key1"], [0])
self.assertEqual(cache["key2"], [1, 2])
self.assertEqual(cache["key3"], [3])
self.assertEqual(cache["key4"], [4])
self.assertEqual(len(cache), 5)
cache["key5"] = [5, 6]
self.assertEqual(len(cache), 4)
self.assertEqual(cache.get("key1"), None)
self.assertEqual(cache.get("key2"), None)
self.assertEqual(cache["key3"], [3])
self.assertEqual(cache["key4"], [4])
self.assertEqual(cache["key5"], [5, 6])
def test_zero_size_drop_from_cache(self) -> None:
"""Test that `drop_from_cache` works correctly with 0-sized entries."""
cache: LruCache[str, list[int]] = LruCache(
max_size=5,
clock=self.clock,
size_callback=lambda x: 0,
server_name="test_server",
)
cache["key1"] = []
self.assertEqual(len(cache), 0)
assert isinstance(cache.cache, dict)
cache.cache["key1"].drop_from_cache()
self.assertIsNone(
cache.pop("key1"), "Cache entry should have been evicted but wasn't"
)
class TimeEvictionTestCase(unittest.HomeserverTestCase):
"""Test that time based eviction works correctly."""
def default_config(self) -> JsonDict:
config = super().default_config()
config.setdefault("caches", {})["expiry_time"] = "30m"
return config
def test_evict(self) -> None:
setup_expire_lru_cache_entries(self.hs)
cache: LruCache[str, int] = LruCache(
max_size=5, server_name="test_server", clock=self.hs.get_clock()
)
# Check that we evict entries we haven't accessed for 30 minutes.
cache["key1"] = 1
cache["key2"] = 2
self.reactor.advance(20 * 60)
self.assertEqual(cache.get("key1"), 1)
self.reactor.advance(20 * 60)
# We have only touched `key1` in the last 30m, so we expect that to
# still be in the cache while `key2` should have been evicted.
self.assertEqual(cache.get("key1"), 1)
self.assertEqual(cache.get("key2"), None)
# Check that re-adding an expired key works correctly.
cache["key2"] = 3
self.assertEqual(cache.get("key2"), 3)
self.reactor.advance(20 * 60)
self.assertEqual(cache.get("key2"), 3)
self.reactor.advance(20 * 60)
self.assertEqual(cache.get("key1"), None)
self.assertEqual(cache.get("key2"), 3)
class MemoryEvictionTestCase(unittest.HomeserverTestCase):
@override_config(
{
"caches": {
"cache_autotuning": {
"max_cache_memory_usage": "700M",
"target_cache_memory_usage": "500M",
"min_cache_ttl": "5m",
}
}
}
)
@patch("synapse.util.caches.lrucache.get_jemalloc_stats")
def test_evict_memory(self, jemalloc_interface: Mock) -> None:
mock_jemalloc_class = Mock(spec=JemallocStats)
jemalloc_interface.return_value = mock_jemalloc_class
# set the return value of get_stat() to be greater than max_cache_memory_usage
mock_jemalloc_class.get_stat.return_value = 924288000
setup_expire_lru_cache_entries(self.hs)
cache: LruCache[str, int] = LruCache(
max_size=4, server_name="test_server", clock=self.hs.get_clock()
)
cache["key1"] = 1
cache["key2"] = 2
# advance the reactor less than the min_cache_ttl
self.reactor.advance(60 * 2)
# our items should still be in the cache
self.assertEqual(cache.get("key1"), 1)
self.assertEqual(cache.get("key2"), 2)
# advance the reactor past the min_cache_ttl
self.reactor.advance(60 * 6)
# the items should be cleared from cache
self.assertEqual(cache.get("key1"), None)
self.assertEqual(cache.get("key2"), None)
# add more stuff to caches
cache["key1"] = 1
cache["key2"] = 2
# set the return value of get_stat() to be lower than target_cache_memory_usage
mock_jemalloc_class.get_stat.return_value = 10000
# advance the reactor past the min_cache_ttl
self.reactor.advance(60 * 6)
# the items should still be in the cache
self.assertEqual(cache.get("key1"), 1)
self.assertEqual(cache.get("key2"), 2)
def _fill_rust_cache(self) -> None:
"""Put one entry in a Rust cache, stamped with the current time."""
user_id = self.get_success(
self.hs.get_registration_handler().register_user(localpart="rust")
)
self.get_success(
self.hs.get_datastores().main.is_feature_enabled(
user_id, ExperimentalFeature.MSC3881
)
)
@override_config(
{
"caches": {
"cache_autotuning": {
"max_cache_memory_usage": "700M",
"target_cache_memory_usage": "500M",
"min_cache_ttl": "5m",
}
}
}
)
@patch("synapse.util.caches.lrucache.get_jemalloc_stats")
def test_evict_memory_reaches_rust_caches(self, jemalloc_interface: Mock) -> None:
"""Rust cache entries are evicted on memory pressure, in order of
last access together with the Python ones, and not before
`min_cache_ttl`."""
mock_jemalloc_class = Mock(spec=JemallocStats)
jemalloc_interface.return_value = mock_jemalloc_class
mock_jemalloc_class.get_stat.return_value = 924288000
rust_cache = self.hs.get_datastores().main.is_feature_enabled_for_user
setup_expire_lru_cache_entries(self.hs)
cache: LruCache[str, int] = LruCache(
max_size=4, server_name="test_server", clock=self.hs.get_clock()
)
cache["old"] = 1
self.reactor.advance(120)
self._fill_rust_cache()
self.assertEqual(len(rust_cache), 1)
self.reactor.advance(120)
cache["new"] = 2
# Past `min_cache_ttl` for the Python entry read first. The Rust entry
# is younger, so it stays.
self.reactor.advance(120)
self.assertNotIn("old", cache)
self.assertEqual(len(rust_cache), 1)
self.assertIn("new", cache)
# Then the Rust entry...
self.reactor.advance(120)
self.assertEqual(len(rust_cache), 0)
self.assertIn("new", cache)
# ...then the Python entry read last.
self.reactor.advance(120)
self.assertNotIn("new", cache)
@override_config(
{
"caches": {
"cache_autotuning": {
"max_cache_memory_usage": "700M",
"target_cache_memory_usage": "500M",
"min_cache_ttl": "5m",
}
}
}
)
@patch("synapse.util.caches.lrucache.get_jemalloc_stats")
def test_evict_memory_stops_after_rust_batch(
self, jemalloc_interface: Mock
) -> None:
"""Memory is checked after each batch of Rust evictions, and once it
is under target, younger Python entries stay."""
mock_jemalloc_class = Mock(spec=JemallocStats)
jemalloc_interface.return_value = mock_jemalloc_class
mock_jemalloc_class.get_stat.return_value = 10000
rust_cache = self.hs.get_datastores().main.is_feature_enabled_for_user
setup_expire_lru_cache_entries(self.hs)
cache: LruCache[str, int] = LruCache(
max_size=4, server_name="test_server", clock=self.hs.get_clock()
)
cache["old"] = 1
self.reactor.advance(60)
self._fill_rust_cache()
self.reactor.advance(60)
cache["new"] = 2
# Everything is old enough to evict, but memory is fine.
self.reactor.advance(600)
self.assertIn("old", cache)
self.assertEqual(len(rust_cache), 1)
self.assertIn("new", cache)
# Memory is high on the next tick's first reading, and back under
# target once the Rust entry has gone.
mock_jemalloc_class.get_stat.side_effect = [924288000] + [10000] * 100
self.reactor.advance(30)
self.assertNotIn("old", cache)
self.assertEqual(len(rust_cache), 0)
self.assertIn("new", cache)
@override_config(
{
"caches": {
"cache_autotuning": {
"max_cache_memory_usage": "700M",
"target_cache_memory_usage": "500M",
"min_cache_ttl": "5m",
}
}
}
)
@patch("synapse.util.caches.lrucache.get_jemalloc_stats")
def test_evict_memory_drains_a_big_rust_cache(
self, jemalloc_interface: Mock
) -> None:
"""Draining more Rust entries than the walk does between yields still
empties the cache, yielding to the reactor on the way."""
mock_jemalloc_class = Mock(spec=JemallocStats)
jemalloc_interface.return_value = mock_jemalloc_class
mock_jemalloc_class.get_stat.return_value = 924288000
setup_expire_lru_cache_entries(self.hs)
cache = RustEventCache(self.hs.get_rust_runtime(), 30000)
event = make_event_from_dict(
{
"type": "m.room.message",
"room_id": "!room:test",
"sender": "@alice:test",
"content": {},
"auth_events": [],
"prev_events": [],
"depth": 1,
"origin_server_ts": 1,
"hashes": {"sha256": "x"},
"signatures": {},
},
RoomVersions.V10,
)
for i in range(25000):
cache.set_local((f"$event{i}",), event, None)
self.assertEqual(len(cache), 25000)
self.reactor.advance(60 * 6)
self.assertEqual(len(cache), 0)
class ExtraIndexLruCacheTestCase(unittest.HomeserverTestCase):
def test_invalidate_simple(self) -> None:
cache: LruCache[str, int] = LruCache(
max_size=10,
clock=self.hs.get_clock(),
server_name="test_server",
extra_index_cb=lambda k, v: str(v),
)
cache["key1"] = 1
cache["key2"] = 2
cache.invalidate_on_extra_index("key1")
self.assertEqual(cache.get("key1"), 1)
self.assertEqual(cache.get("key2"), 2)
cache.invalidate_on_extra_index("1")
self.assertEqual(cache.get("key1"), None)
self.assertEqual(cache.get("key2"), 2)
def test_invalidate_multi(self) -> None:
cache: LruCache[str, int] = LruCache(
max_size=10,
clock=self.hs.get_clock(),
server_name="test_server",
extra_index_cb=lambda k, v: str(v),
)
cache["key1"] = 1
cache["key2"] = 1
cache["key3"] = 2
cache.invalidate_on_extra_index("key1")
self.assertEqual(cache.get("key1"), 1)
self.assertEqual(cache.get("key2"), 1)
self.assertEqual(cache.get("key3"), 2)
cache.invalidate_on_extra_index("1")
self.assertEqual(cache.get("key1"), None)
self.assertEqual(cache.get("key2"), None)
self.assertEqual(cache.get("key3"), 2)
Reference in new issue View Git Blame Copy Permalink