MatrixSynapse/synapse/util/caches/lrucache.py

307 lines
9.7 KiB
Python
Raw Normal View History

2015-02-11 15:52:23 +01:00
# -*- coding: utf-8 -*-
2016-01-07 05:26:29 +01:00
# Copyright 2015, 2016 OpenMarket Ltd
2015-02-11 15:52:23 +01:00
#
# 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 threading
2018-07-09 08:09:20 +02:00
from functools import wraps
from typing import Callable, Optional, Type, Union
from synapse.config import cache as cache_config
from synapse.util.caches.treecache import TreeCache
def enumerate_leaves(node, depth):
if depth == 0:
yield node
else:
for n in node.values():
for m in enumerate_leaves(n, depth - 1):
yield m
2020-09-04 12:54:56 +02:00
class _Node:
2016-08-19 12:18:26 +02:00
__slots__ = ["prev_node", "next_node", "key", "value", "callbacks"]
def __init__(self, prev_node, next_node, key, value, callbacks=set()):
self.prev_node = prev_node
self.next_node = next_node
self.key = key
self.value = value
2016-08-19 12:18:26 +02:00
self.callbacks = callbacks
2020-09-04 12:54:56 +02:00
class LruCache:
2016-01-22 13:21:13 +01:00
"""
Least-recently-used cache.
Supports del_multi only if cache_type=TreeCache
If cache_type=TreeCache, all keys must be tuples.
2016-08-19 12:18:26 +02:00
Can also set callbacks on objects when getting/setting which are fired
when that key gets invalidated/evicted.
2016-01-22 13:21:13 +01:00
"""
2019-06-20 11:32:02 +02:00
def __init__(
self,
max_size: int,
keylen: int = 1,
cache_type: Type[Union[dict, TreeCache]] = dict,
size_callback: Optional[Callable] = None,
evicted_callback: Optional[Callable] = None,
apply_cache_factor_from_config: bool = True,
2019-06-20 11:32:02 +02:00
):
"""
Args:
max_size: The maximum amount of entries the cache can hold
keylen: The length of the tuple used as the cache key
cache_type (type):
type of underlying cache to be used. Typically one of dict
or TreeCache.
size_callback (func(V) -> int | None):
evicted_callback (func(int)|None):
if not None, called on eviction with the size of the evicted
entry
apply_cache_factor_from_config (bool): If true, `max_size` will be
multiplied by a cache factor derived from the homeserver config
"""
cache = cache_type()
2016-01-29 11:00:45 +01:00
self.cache = cache # Used for introspection.
self.apply_cache_factor_from_config = apply_cache_factor_from_config
# Save the original max size, and apply the default size factor.
self._original_max_size = max_size
# We previously didn't apply the cache factor here, and as such some caches were
# not affected by the global cache factor. Add an option here to disable applying
# the cache factor when a cache is created
if apply_cache_factor_from_config:
self.max_size = int(max_size * cache_config.properties.default_factor_size)
else:
self.max_size = int(max_size)
list_root = _Node(None, None, None, None)
list_root.next_node = list_root
list_root.prev_node = list_root
2015-02-11 15:52:23 +01:00
lock = threading.Lock()
def evict():
while cache_len() > self.max_size:
todelete = list_root.prev_node
evicted_len = delete_node(todelete)
cache.pop(todelete.key, None)
if evicted_callback:
evicted_callback(evicted_len)
def synchronized(f):
@wraps(f)
def inner(*args, **kwargs):
with lock:
return f(*args, **kwargs)
return inner
cached_cache_len = [0]
if size_callback is not None:
2019-06-20 11:32:02 +02:00
def cache_len():
return cached_cache_len[0]
2019-06-20 11:32:02 +02:00
else:
2019-06-20 11:32:02 +02:00
def cache_len():
return len(cache)
self.len = synchronized(cache_len)
def add_node(key, value, callbacks=set()):
2015-02-11 15:52:23 +01:00
prev_node = list_root
next_node = prev_node.next_node
2016-08-19 12:18:26 +02:00
node = _Node(prev_node, next_node, key, value, callbacks)
prev_node.next_node = node
next_node.prev_node = node
2015-02-11 15:52:23 +01:00
cache[key] = node
if size_callback:
cached_cache_len[0] += size_callback(node.value)
2015-02-11 15:52:23 +01:00
def move_node_to_front(node):
prev_node = node.prev_node
next_node = node.next_node
prev_node.next_node = next_node
next_node.prev_node = prev_node
2015-02-11 15:52:23 +01:00
prev_node = list_root
next_node = prev_node.next_node
node.prev_node = prev_node
node.next_node = next_node
prev_node.next_node = node
next_node.prev_node = node
2015-02-11 15:52:23 +01:00
def delete_node(node):
prev_node = node.prev_node
next_node = node.next_node
prev_node.next_node = next_node
next_node.prev_node = prev_node
2015-02-11 15:52:23 +01:00
deleted_len = 1
if size_callback:
deleted_len = size_callback(node.value)
cached_cache_len[0] -= deleted_len
2016-08-19 12:18:26 +02:00
for cb in node.callbacks:
cb()
node.callbacks.clear()
return deleted_len
2016-08-19 12:18:26 +02:00
@synchronized
def cache_get(key, default=None, callbacks=[]):
2015-02-11 15:52:23 +01:00
node = cache.get(key, None)
if node is not None:
move_node_to_front(node)
node.callbacks.update(callbacks)
return node.value
2015-02-11 15:52:23 +01:00
else:
return default
@synchronized
def cache_set(key, value, callbacks=[]):
2015-02-11 15:52:23 +01:00
node = cache.get(key, None)
if node is not None:
2018-03-19 12:35:53 +01:00
# We sometimes store large objects, e.g. dicts, which cause
# the inequality check to take a long time. So let's only do
# the check if we have some callbacks to call.
if node.callbacks and value != node.value:
2016-08-19 12:18:26 +02:00
for cb in node.callbacks:
cb()
node.callbacks.clear()
2016-08-19 12:18:26 +02:00
2018-03-19 12:35:53 +01:00
# We don't bother to protect this by value != node.value as
# generally size_callback will be cheap compared with equality
# checks. (For example, taking the size of two dicts is quicker
# than comparing them for equality.)
if size_callback:
cached_cache_len[0] -= size_callback(node.value)
cached_cache_len[0] += size_callback(value)
node.callbacks.update(callbacks)
2016-08-19 12:18:26 +02:00
2015-02-11 15:52:23 +01:00
move_node_to_front(node)
node.value = value
2015-02-11 15:52:23 +01:00
else:
add_node(key, value, set(callbacks))
evict()
2015-02-11 15:52:23 +01:00
@synchronized
2015-02-11 15:52:23 +01:00
def cache_set_default(key, value):
node = cache.get(key, None)
if node is not None:
return node.value
2015-02-11 15:52:23 +01:00
else:
add_node(key, value)
evict()
2015-02-11 15:52:23 +01:00
return value
@synchronized
2015-02-11 15:52:23 +01:00
def cache_pop(key, default=None):
node = cache.get(key, None)
if node:
delete_node(node)
cache.pop(node.key, None)
return node.value
2015-02-11 15:52:23 +01:00
else:
return default
@synchronized
def cache_del_multi(key):
"""
This will only work if constructed with cache_type=TreeCache
"""
popped = cache.pop(key)
if popped is None:
return
for leaf in enumerate_leaves(popped, keylen - len(key)):
delete_node(leaf)
@synchronized
def cache_clear():
list_root.next_node = list_root
list_root.prev_node = list_root
2016-08-19 12:18:26 +02:00
for node in cache.values():
for cb in node.callbacks:
cb()
cache.clear()
2017-01-18 15:55:23 +01:00
if size_callback:
cached_cache_len[0] = 0
@synchronized
def cache_contains(key):
return key in cache
2015-02-11 15:52:23 +01:00
self.sentinel = object()
self._on_resize = evict
2015-02-11 15:52:23 +01:00
self.get = cache_get
self.set = cache_set
self.setdefault = cache_set_default
self.pop = cache_pop
if cache_type is TreeCache:
self.del_multi = cache_del_multi
self.len = synchronized(cache_len)
self.contains = cache_contains
self.clear = cache_clear
2015-02-11 15:52:23 +01:00
def __getitem__(self, key):
result = self.get(key, self.sentinel)
if result is self.sentinel:
raise KeyError()
else:
return result
def __setitem__(self, key, value):
self.set(key, value)
def __delitem__(self, key, value):
result = self.pop(key, self.sentinel)
if result is self.sentinel:
raise KeyError()
def __len__(self):
return self.len()
def __contains__(self, key):
return self.contains(key)
def set_cache_factor(self, factor: float) -> bool:
"""
Set the cache factor for this individual cache.
This will trigger a resize if it changes, which may require evicting
items from the cache.
Returns:
bool: Whether the cache changed size or not.
"""
if not self.apply_cache_factor_from_config:
return False
new_size = int(self._original_max_size * factor)
if new_size != self.max_size:
self.max_size = new_size
self._on_resize()
return True
return False