2015-08-11 12:59:32 -04:00
|
|
|
# -*- coding: utf-8 -*-
|
2016-01-06 23:26:29 -05:00
|
|
|
# Copyright 2015, 2016 OpenMarket Ltd
|
2015-08-11 12:59:32 -04:00
|
|
|
#
|
|
|
|
# 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 logging
|
|
|
|
|
|
|
|
from synapse.util.async import ObservableDeferred
|
|
|
|
from synapse.util import unwrapFirstError
|
|
|
|
from synapse.util.caches.lrucache import LruCache
|
2016-01-22 07:10:33 -05:00
|
|
|
from synapse.util.caches.treecache import TreeCache
|
2016-02-04 05:22:44 -05:00
|
|
|
from synapse.util.logcontext import (
|
|
|
|
PreserveLoggingContext, preserve_context_over_deferred, preserve_context_over_fn
|
|
|
|
)
|
2015-08-12 05:13:35 -04:00
|
|
|
|
|
|
|
from . import caches_by_name, DEBUG_CACHES, cache_counter
|
2015-08-11 12:59:32 -04:00
|
|
|
|
|
|
|
from twisted.internet import defer
|
|
|
|
|
|
|
|
from collections import OrderedDict
|
|
|
|
|
2016-03-01 07:56:39 -05:00
|
|
|
import os
|
2015-08-11 12:59:32 -04:00
|
|
|
import functools
|
|
|
|
import inspect
|
|
|
|
import threading
|
2016-06-03 06:47:07 -04:00
|
|
|
|
2015-08-11 12:59:32 -04:00
|
|
|
|
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
|
|
|
|
|
|
|
|
_CacheSentinel = object()
|
|
|
|
|
|
|
|
|
2016-03-01 07:56:39 -05:00
|
|
|
CACHE_SIZE_FACTOR = float(os.environ.get("SYNAPSE_CACHE_FACTOR", 0.1))
|
|
|
|
|
|
|
|
|
2015-08-11 12:59:32 -04:00
|
|
|
class Cache(object):
|
2016-06-01 13:01:22 -04:00
|
|
|
__slots__ = (
|
|
|
|
"cache",
|
|
|
|
"max_entries",
|
|
|
|
"name",
|
|
|
|
"keylen",
|
|
|
|
"sequence",
|
|
|
|
"thread",
|
|
|
|
)
|
2015-08-11 12:59:32 -04:00
|
|
|
|
2016-01-22 07:10:33 -05:00
|
|
|
def __init__(self, name, max_entries=1000, keylen=1, lru=True, tree=False):
|
2015-08-11 12:59:32 -04:00
|
|
|
if lru:
|
2016-01-22 07:10:33 -05:00
|
|
|
cache_type = TreeCache if tree else dict
|
|
|
|
self.cache = LruCache(
|
|
|
|
max_size=max_entries, keylen=keylen, cache_type=cache_type
|
|
|
|
)
|
2015-08-11 12:59:32 -04:00
|
|
|
self.max_entries = None
|
|
|
|
else:
|
|
|
|
self.cache = OrderedDict()
|
|
|
|
self.max_entries = max_entries
|
|
|
|
|
|
|
|
self.name = name
|
|
|
|
self.keylen = keylen
|
|
|
|
self.sequence = 0
|
|
|
|
self.thread = None
|
|
|
|
caches_by_name[name] = self.cache
|
|
|
|
|
|
|
|
def check_thread(self):
|
|
|
|
expected_thread = self.thread
|
|
|
|
if expected_thread is None:
|
|
|
|
self.thread = threading.current_thread()
|
|
|
|
else:
|
|
|
|
if expected_thread is not threading.current_thread():
|
|
|
|
raise ValueError(
|
|
|
|
"Cache objects can only be accessed from the main thread"
|
|
|
|
)
|
|
|
|
|
|
|
|
def get(self, key, default=_CacheSentinel):
|
|
|
|
val = self.cache.get(key, _CacheSentinel)
|
|
|
|
if val is not _CacheSentinel:
|
|
|
|
cache_counter.inc_hits(self.name)
|
|
|
|
return val
|
|
|
|
|
|
|
|
cache_counter.inc_misses(self.name)
|
|
|
|
|
|
|
|
if default is _CacheSentinel:
|
|
|
|
raise KeyError()
|
|
|
|
else:
|
|
|
|
return default
|
|
|
|
|
|
|
|
def update(self, sequence, key, value):
|
|
|
|
self.check_thread()
|
|
|
|
if self.sequence == sequence:
|
|
|
|
# Only update the cache if the caches sequence number matches the
|
|
|
|
# number that the cache had before the SELECT was started (SYN-369)
|
|
|
|
self.prefill(key, value)
|
|
|
|
|
|
|
|
def prefill(self, key, value):
|
|
|
|
if self.max_entries is not None:
|
|
|
|
while len(self.cache) >= self.max_entries:
|
|
|
|
self.cache.popitem(last=False)
|
|
|
|
|
|
|
|
self.cache[key] = value
|
|
|
|
|
|
|
|
def invalidate(self, key):
|
|
|
|
self.check_thread()
|
|
|
|
if not isinstance(key, tuple):
|
2015-08-12 05:30:30 -04:00
|
|
|
raise TypeError(
|
|
|
|
"The cache key must be a tuple not %r" % (type(key),)
|
|
|
|
)
|
2015-08-11 12:59:32 -04:00
|
|
|
|
|
|
|
# Increment the sequence number so that any SELECT statements that
|
|
|
|
# raced with the INSERT don't update the cache (SYN-369)
|
|
|
|
self.sequence += 1
|
|
|
|
self.cache.pop(key, None)
|
|
|
|
|
2016-01-21 14:16:25 -05:00
|
|
|
def invalidate_many(self, key):
|
|
|
|
self.check_thread()
|
|
|
|
if not isinstance(key, tuple):
|
|
|
|
raise TypeError(
|
|
|
|
"The cache key must be a tuple not %r" % (type(key),)
|
|
|
|
)
|
|
|
|
self.sequence += 1
|
|
|
|
self.cache.del_multi(key)
|
|
|
|
|
2015-08-11 12:59:32 -04:00
|
|
|
def invalidate_all(self):
|
|
|
|
self.check_thread()
|
|
|
|
self.sequence += 1
|
|
|
|
self.cache.clear()
|
|
|
|
|
|
|
|
|
|
|
|
class CacheDescriptor(object):
|
|
|
|
""" A method decorator that applies a memoizing cache around the function.
|
|
|
|
|
|
|
|
This caches deferreds, rather than the results themselves. Deferreds that
|
|
|
|
fail are removed from the cache.
|
|
|
|
|
|
|
|
The function is presumed to take zero or more arguments, which are used in
|
|
|
|
a tuple as the key for the cache. Hits are served directly from the cache;
|
|
|
|
misses use the function body to generate the value.
|
|
|
|
|
|
|
|
The wrapped function has an additional member, a callable called
|
|
|
|
"invalidate". This can be used to remove individual entries from the cache.
|
|
|
|
|
|
|
|
The wrapped function has another additional callable, called "prefill",
|
|
|
|
which can be used to insert values into the cache specifically, without
|
|
|
|
calling the calculation function.
|
|
|
|
"""
|
2016-01-22 07:10:33 -05:00
|
|
|
def __init__(self, orig, max_entries=1000, num_args=1, lru=True, tree=False,
|
2015-08-11 12:59:32 -04:00
|
|
|
inlineCallbacks=False):
|
2016-03-01 08:21:46 -05:00
|
|
|
max_entries = int(max_entries * CACHE_SIZE_FACTOR)
|
|
|
|
|
2015-08-11 12:59:32 -04:00
|
|
|
self.orig = orig
|
|
|
|
|
|
|
|
if inlineCallbacks:
|
|
|
|
self.function_to_call = defer.inlineCallbacks(orig)
|
|
|
|
else:
|
|
|
|
self.function_to_call = orig
|
|
|
|
|
|
|
|
self.max_entries = max_entries
|
|
|
|
self.num_args = num_args
|
|
|
|
self.lru = lru
|
2016-01-22 07:10:33 -05:00
|
|
|
self.tree = tree
|
2015-08-11 12:59:32 -04:00
|
|
|
|
2016-02-02 12:18:50 -05:00
|
|
|
self.arg_names = inspect.getargspec(orig).args[1:num_args + 1]
|
2015-08-11 12:59:32 -04:00
|
|
|
|
|
|
|
if len(self.arg_names) < self.num_args:
|
|
|
|
raise Exception(
|
|
|
|
"Not enough explicit positional arguments to key off of for %r."
|
|
|
|
" (@cached cannot key off of *args or **kwars)"
|
|
|
|
% (orig.__name__,)
|
|
|
|
)
|
|
|
|
|
2016-04-06 08:08:05 -04:00
|
|
|
def __get__(self, obj, objtype=None):
|
|
|
|
cache = Cache(
|
2015-08-11 12:59:32 -04:00
|
|
|
name=self.orig.__name__,
|
|
|
|
max_entries=self.max_entries,
|
|
|
|
keylen=self.num_args,
|
|
|
|
lru=self.lru,
|
2016-01-22 07:10:33 -05:00
|
|
|
tree=self.tree,
|
2015-08-11 12:59:32 -04:00
|
|
|
)
|
|
|
|
|
|
|
|
@functools.wraps(self.orig)
|
|
|
|
def wrapped(*args, **kwargs):
|
|
|
|
arg_dict = inspect.getcallargs(self.orig, obj, *args, **kwargs)
|
|
|
|
cache_key = tuple(arg_dict[arg_nm] for arg_nm in self.arg_names)
|
|
|
|
try:
|
2016-04-06 08:08:05 -04:00
|
|
|
cached_result_d = cache.get(cache_key)
|
2015-08-11 12:59:32 -04:00
|
|
|
|
|
|
|
observer = cached_result_d.observe()
|
|
|
|
if DEBUG_CACHES:
|
|
|
|
@defer.inlineCallbacks
|
|
|
|
def check_result(cached_result):
|
|
|
|
actual_result = yield self.function_to_call(obj, *args, **kwargs)
|
|
|
|
if actual_result != cached_result:
|
|
|
|
logger.error(
|
|
|
|
"Stale cache entry %s%r: cached: %r, actual %r",
|
|
|
|
self.orig.__name__, cache_key,
|
|
|
|
cached_result, actual_result,
|
|
|
|
)
|
|
|
|
raise ValueError("Stale cache entry")
|
|
|
|
defer.returnValue(cached_result)
|
|
|
|
observer.addCallback(check_result)
|
|
|
|
|
2016-02-04 05:22:44 -05:00
|
|
|
return preserve_context_over_deferred(observer)
|
2015-08-11 12:59:32 -04:00
|
|
|
except KeyError:
|
|
|
|
# Get the sequence number of the cache before reading from the
|
|
|
|
# database so that we can tell if the cache is invalidated
|
|
|
|
# while the SELECT is executing (SYN-369)
|
2016-04-06 08:08:05 -04:00
|
|
|
sequence = cache.sequence
|
2015-08-11 12:59:32 -04:00
|
|
|
|
|
|
|
ret = defer.maybeDeferred(
|
2016-02-04 05:22:44 -05:00
|
|
|
preserve_context_over_fn,
|
2015-08-11 12:59:32 -04:00
|
|
|
self.function_to_call,
|
|
|
|
obj, *args, **kwargs
|
|
|
|
)
|
|
|
|
|
|
|
|
def onErr(f):
|
2016-04-06 08:08:05 -04:00
|
|
|
cache.invalidate(cache_key)
|
2015-08-11 12:59:32 -04:00
|
|
|
return f
|
|
|
|
|
|
|
|
ret.addErrback(onErr)
|
|
|
|
|
|
|
|
ret = ObservableDeferred(ret, consumeErrors=True)
|
2016-04-06 08:08:05 -04:00
|
|
|
cache.update(sequence, cache_key, ret)
|
2015-08-11 12:59:32 -04:00
|
|
|
|
2016-02-04 05:22:44 -05:00
|
|
|
return preserve_context_over_deferred(ret.observe())
|
2015-08-11 12:59:32 -04:00
|
|
|
|
2016-04-06 08:08:05 -04:00
|
|
|
wrapped.invalidate = cache.invalidate
|
|
|
|
wrapped.invalidate_all = cache.invalidate_all
|
|
|
|
wrapped.invalidate_many = cache.invalidate_many
|
|
|
|
wrapped.prefill = cache.prefill
|
|
|
|
wrapped.cache = cache
|
2015-08-11 12:59:32 -04:00
|
|
|
|
|
|
|
obj.__dict__[self.orig.__name__] = wrapped
|
|
|
|
|
|
|
|
return wrapped
|
|
|
|
|
|
|
|
|
|
|
|
class CacheListDescriptor(object):
|
|
|
|
"""Wraps an existing cache to support bulk fetching of keys.
|
|
|
|
|
|
|
|
Given a list of keys it looks in the cache to find any hits, then passes
|
|
|
|
the list of missing keys to the wrapped fucntion.
|
|
|
|
"""
|
|
|
|
|
2016-04-06 08:08:05 -04:00
|
|
|
def __init__(self, orig, cached_method_name, list_name, num_args=1,
|
|
|
|
inlineCallbacks=False):
|
2015-08-11 12:59:32 -04:00
|
|
|
"""
|
|
|
|
Args:
|
|
|
|
orig (function)
|
2016-04-06 08:08:05 -04:00
|
|
|
method_name (str); The name of the chached method.
|
2015-08-11 12:59:32 -04:00
|
|
|
list_name (str): Name of the argument which is the bulk lookup list
|
|
|
|
num_args (int)
|
|
|
|
inlineCallbacks (bool): Whether orig is a generator that should
|
|
|
|
be wrapped by defer.inlineCallbacks
|
|
|
|
"""
|
|
|
|
self.orig = orig
|
|
|
|
|
|
|
|
if inlineCallbacks:
|
|
|
|
self.function_to_call = defer.inlineCallbacks(orig)
|
|
|
|
else:
|
|
|
|
self.function_to_call = orig
|
|
|
|
|
|
|
|
self.num_args = num_args
|
|
|
|
self.list_name = list_name
|
|
|
|
|
2016-02-02 12:18:50 -05:00
|
|
|
self.arg_names = inspect.getargspec(orig).args[1:num_args + 1]
|
2015-08-11 12:59:32 -04:00
|
|
|
self.list_pos = self.arg_names.index(self.list_name)
|
|
|
|
|
2016-04-06 08:08:05 -04:00
|
|
|
self.cached_method_name = cached_method_name
|
2015-08-11 12:59:32 -04:00
|
|
|
|
|
|
|
self.sentinel = object()
|
|
|
|
|
|
|
|
if len(self.arg_names) < self.num_args:
|
|
|
|
raise Exception(
|
|
|
|
"Not enough explicit positional arguments to key off of for %r."
|
|
|
|
" (@cached cannot key off of *args or **kwars)"
|
|
|
|
% (orig.__name__,)
|
|
|
|
)
|
|
|
|
|
|
|
|
if self.list_name not in self.arg_names:
|
|
|
|
raise Exception(
|
|
|
|
"Couldn't see arguments %r for %r."
|
2016-04-06 08:08:05 -04:00
|
|
|
% (self.list_name, cached_method_name,)
|
2015-08-11 12:59:32 -04:00
|
|
|
)
|
|
|
|
|
|
|
|
def __get__(self, obj, objtype=None):
|
|
|
|
|
2016-04-06 08:08:05 -04:00
|
|
|
cache = getattr(obj, self.cached_method_name).cache
|
|
|
|
|
2015-08-11 12:59:32 -04:00
|
|
|
@functools.wraps(self.orig)
|
|
|
|
def wrapped(*args, **kwargs):
|
|
|
|
arg_dict = inspect.getcallargs(self.orig, obj, *args, **kwargs)
|
|
|
|
keyargs = [arg_dict[arg_nm] for arg_nm in self.arg_names]
|
|
|
|
list_args = arg_dict[self.list_name]
|
|
|
|
|
|
|
|
# cached is a dict arg -> deferred, where deferred results in a
|
|
|
|
# 2-tuple (`arg`, `result`)
|
2016-06-01 13:01:22 -04:00
|
|
|
results = {}
|
|
|
|
cached_defers = {}
|
2015-08-11 12:59:32 -04:00
|
|
|
missing = []
|
|
|
|
for arg in list_args:
|
|
|
|
key = list(keyargs)
|
|
|
|
key[self.list_pos] = arg
|
|
|
|
|
|
|
|
try:
|
2016-06-01 13:01:22 -04:00
|
|
|
res = cache.get(tuple(key))
|
2016-06-02 06:52:32 -04:00
|
|
|
if not res.has_succeeded():
|
2016-06-01 13:01:22 -04:00
|
|
|
res = res.observe()
|
|
|
|
res.addCallback(lambda r, arg: (arg, r), arg)
|
|
|
|
cached_defers[arg] = res
|
|
|
|
else:
|
2016-06-02 06:52:32 -04:00
|
|
|
results[arg] = res.get_result()
|
2015-08-11 12:59:32 -04:00
|
|
|
except KeyError:
|
|
|
|
missing.append(arg)
|
|
|
|
|
|
|
|
if missing:
|
2016-04-06 08:08:05 -04:00
|
|
|
sequence = cache.sequence
|
2015-08-11 12:59:32 -04:00
|
|
|
args_to_call = dict(arg_dict)
|
|
|
|
args_to_call[self.list_name] = missing
|
|
|
|
|
|
|
|
ret_d = defer.maybeDeferred(
|
2016-02-04 05:22:44 -05:00
|
|
|
preserve_context_over_fn,
|
2015-08-11 12:59:32 -04:00
|
|
|
self.function_to_call,
|
|
|
|
**args_to_call
|
|
|
|
)
|
|
|
|
|
|
|
|
ret_d = ObservableDeferred(ret_d)
|
|
|
|
|
|
|
|
# We need to create deferreds for each arg in the list so that
|
|
|
|
# we can insert the new deferred into the cache.
|
|
|
|
for arg in missing:
|
2016-02-04 05:22:44 -05:00
|
|
|
with PreserveLoggingContext():
|
|
|
|
observer = ret_d.observe()
|
2015-08-18 06:11:33 -04:00
|
|
|
observer.addCallback(lambda r, arg: r.get(arg, None), arg)
|
2015-08-11 12:59:32 -04:00
|
|
|
|
|
|
|
observer = ObservableDeferred(observer)
|
|
|
|
|
|
|
|
key = list(keyargs)
|
|
|
|
key[self.list_pos] = arg
|
2016-04-06 08:08:05 -04:00
|
|
|
cache.update(sequence, tuple(key), observer)
|
2015-08-11 12:59:32 -04:00
|
|
|
|
|
|
|
def invalidate(f, key):
|
2016-04-06 08:08:05 -04:00
|
|
|
cache.invalidate(key)
|
2015-08-11 12:59:32 -04:00
|
|
|
return f
|
|
|
|
observer.addErrback(invalidate, tuple(key))
|
|
|
|
|
|
|
|
res = observer.observe()
|
|
|
|
res.addCallback(lambda r, arg: (arg, r), arg)
|
|
|
|
|
2016-06-01 13:01:22 -04:00
|
|
|
cached_defers[arg] = res
|
|
|
|
|
|
|
|
if cached_defers:
|
2016-06-03 06:47:07 -04:00
|
|
|
def update_results_dict(res):
|
|
|
|
results.update(res)
|
|
|
|
return results
|
|
|
|
|
2016-06-01 13:01:22 -04:00
|
|
|
return preserve_context_over_deferred(defer.gatherResults(
|
|
|
|
cached_defers.values(),
|
|
|
|
consumeErrors=True,
|
2016-06-03 06:47:07 -04:00
|
|
|
).addCallback(update_results_dict).addErrback(
|
2016-06-01 13:01:22 -04:00
|
|
|
unwrapFirstError
|
2016-06-03 06:47:07 -04:00
|
|
|
))
|
2016-06-01 13:01:22 -04:00
|
|
|
else:
|
|
|
|
return results
|
2015-08-11 12:59:32 -04:00
|
|
|
|
|
|
|
obj.__dict__[self.orig.__name__] = wrapped
|
|
|
|
|
|
|
|
return wrapped
|
|
|
|
|
|
|
|
|
2016-01-22 07:10:33 -05:00
|
|
|
def cached(max_entries=1000, num_args=1, lru=True, tree=False):
|
2015-08-11 12:59:32 -04:00
|
|
|
return lambda orig: CacheDescriptor(
|
|
|
|
orig,
|
|
|
|
max_entries=max_entries,
|
|
|
|
num_args=num_args,
|
2016-01-22 07:10:33 -05:00
|
|
|
lru=lru,
|
|
|
|
tree=tree,
|
2015-08-11 12:59:32 -04:00
|
|
|
)
|
|
|
|
|
|
|
|
|
2016-01-22 07:10:33 -05:00
|
|
|
def cachedInlineCallbacks(max_entries=1000, num_args=1, lru=False, tree=False):
|
2015-08-11 12:59:32 -04:00
|
|
|
return lambda orig: CacheDescriptor(
|
|
|
|
orig,
|
|
|
|
max_entries=max_entries,
|
|
|
|
num_args=num_args,
|
|
|
|
lru=lru,
|
2016-01-22 07:10:33 -05:00
|
|
|
tree=tree,
|
2015-08-11 12:59:32 -04:00
|
|
|
inlineCallbacks=True,
|
|
|
|
)
|
|
|
|
|
|
|
|
|
2016-04-06 08:08:05 -04:00
|
|
|
def cachedList(cached_method_name, list_name, num_args=1, inlineCallbacks=False):
|
2015-08-12 11:42:46 -04:00
|
|
|
"""Creates a descriptor that wraps a function in a `CacheListDescriptor`.
|
|
|
|
|
|
|
|
Used to do batch lookups for an already created cache. A single argument
|
|
|
|
is specified as a list that is iterated through to lookup keys in the
|
|
|
|
original cache. A new list consisting of the keys that weren't in the cache
|
|
|
|
get passed to the original function, the result of which is stored in the
|
|
|
|
cache.
|
|
|
|
|
|
|
|
Args:
|
|
|
|
cache (Cache): The underlying cache to use.
|
|
|
|
list_name (str): The name of the argument that is the list to use to
|
|
|
|
do batch lookups in the cache.
|
|
|
|
num_args (int): Number of arguments to use as the key in the cache.
|
|
|
|
inlineCallbacks (bool): Should the function be wrapped in an
|
|
|
|
`defer.inlineCallbacks`?
|
|
|
|
|
|
|
|
Example:
|
|
|
|
|
|
|
|
class Example(object):
|
|
|
|
@cached(num_args=2)
|
|
|
|
def do_something(self, first_arg):
|
|
|
|
...
|
|
|
|
|
|
|
|
@cachedList(do_something.cache, list_name="second_args", num_args=2)
|
|
|
|
def batch_do_something(self, first_arg, second_args):
|
|
|
|
...
|
|
|
|
"""
|
2015-08-11 12:59:32 -04:00
|
|
|
return lambda orig: CacheListDescriptor(
|
|
|
|
orig,
|
2016-04-06 08:08:05 -04:00
|
|
|
cached_method_name=cached_method_name,
|
2015-08-11 12:59:32 -04:00
|
|
|
list_name=list_name,
|
|
|
|
num_args=num_args,
|
|
|
|
inlineCallbacks=inlineCallbacks,
|
|
|
|
)
|