summary refs log tree commit diff
path: root/synapse/util/caches/descriptors.py
blob: 5d7fffee66ea596e4d21ee066cb6507a0ea2d84f (plain) (blame)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
# -*- coding: utf-8 -*-
# Copyright 2015, 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 functools
import inspect
import logging
from typing import Any, Callable, Generic, Optional, Tuple, TypeVar, Union, cast
from weakref import WeakValueDictionary

from twisted.internet import defer

from synapse.logging.context import make_deferred_yieldable, preserve_fn
from synapse.util import unwrapFirstError
from synapse.util.caches.deferred_cache import DeferredCache

logger = logging.getLogger(__name__)

CacheKey = Union[Tuple, Any]

F = TypeVar("F", bound=Callable[..., Any])


class _CachedFunction(Generic[F]):
    invalidate = None  # type: Any
    invalidate_all = None  # type: Any
    invalidate_many = None  # type: Any
    prefill = None  # type: Any
    cache = None  # type: Any
    num_args = None  # type: Any

    __name__ = None  # type: str

    # Note: This function signature is actually fiddled with by the synapse mypy
    # plugin to a) make it a bound method, and b) remove any `cache_context` arg.
    __call__ = None  # type: F


class _CacheDescriptorBase:
    def __init__(self, orig: _CachedFunction, num_args, cache_context=False):
        self.orig = orig

        arg_spec = inspect.getfullargspec(orig)
        all_args = arg_spec.args

        if "cache_context" in all_args:
            if not cache_context:
                raise ValueError(
                    "Cannot have a 'cache_context' arg without setting"
                    " cache_context=True"
                )
        elif cache_context:
            raise ValueError(
                "Cannot have cache_context=True without having an arg"
                " named `cache_context`"
            )

        if num_args is None:
            num_args = len(all_args) - 1
            if cache_context:
                num_args -= 1

        if len(all_args) < num_args + 1:
            raise Exception(
                "Not enough explicit positional arguments to key off for %r: "
                "got %i args, but wanted %i. (@cached cannot key off *args or "
                "**kwargs)" % (orig.__name__, len(all_args), num_args)
            )

        self.num_args = num_args

        # list of the names of the args used as the cache key
        self.arg_names = all_args[1 : num_args + 1]

        # self.arg_defaults is a map of arg name to its default value for each
        # argument that has a default value
        if arg_spec.defaults:
            self.arg_defaults = dict(
                zip(all_args[-len(arg_spec.defaults) :], arg_spec.defaults)
            )
        else:
            self.arg_defaults = {}

        if "cache_context" in self.arg_names:
            raise Exception("cache_context arg cannot be included among the cache keys")

        self.add_cache_context = cache_context


class CacheDescriptor(_CacheDescriptorBase):
    """ 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.

    Cached functions can be "chained" (i.e. a cached function can call other cached
    functions and get appropriately invalidated when they called caches are
    invalidated) by adding a special "cache_context" argument to the function
    and passing that as a kwarg to all caches called. For example::

        @cached(cache_context=True)
        def foo(self, key, cache_context):
            r1 = yield self.bar1(key, on_invalidate=cache_context.invalidate)
            r2 = yield self.bar2(key, on_invalidate=cache_context.invalidate)
            return r1 + r2

    Args:
        num_args (int): number of positional arguments (excluding ``self`` and
            ``cache_context``) to use as cache keys. Defaults to all named
            args of the function.
    """

    def __init__(
        self,
        orig,
        max_entries=1000,
        num_args=None,
        tree=False,
        cache_context=False,
        iterable=False,
    ):

        super().__init__(orig, num_args=num_args, cache_context=cache_context)

        self.max_entries = max_entries
        self.tree = tree
        self.iterable = iterable

    def __get__(self, obj, owner):
        cache = DeferredCache(
            name=self.orig.__name__,
            max_entries=self.max_entries,
            keylen=self.num_args,
            tree=self.tree,
            iterable=self.iterable,
        )  # type: DeferredCache[CacheKey, Any]

        def get_cache_key_gen(args, kwargs):
            """Given some args/kwargs return a generator that resolves into
            the cache_key.

            We loop through each arg name, looking up if its in the `kwargs`,
            otherwise using the next argument in `args`. If there are no more
            args then we try looking the arg name up in the defaults
            """
            pos = 0
            for nm in self.arg_names:
                if nm in kwargs:
                    yield kwargs[nm]
                elif pos < len(args):
                    yield args[pos]
                    pos += 1
                else:
                    yield self.arg_defaults[nm]

        # By default our cache key is a tuple, but if there is only one item
        # then don't bother wrapping in a tuple.  This is to save memory.
        if self.num_args == 1:
            nm = self.arg_names[0]

            def get_cache_key(args, kwargs):
                if nm in kwargs:
                    return kwargs[nm]
                elif len(args):
                    return args[0]
                else:
                    return self.arg_defaults[nm]

        else:

            def get_cache_key(args, kwargs):
                return tuple(get_cache_key_gen(args, kwargs))

        @functools.wraps(self.orig)
        def _wrapped(*args, **kwargs):
            # If we're passed a cache_context then we'll want to call its invalidate()
            # whenever we are invalidated
            invalidate_callback = kwargs.pop("on_invalidate", None)

            cache_key = get_cache_key(args, kwargs)

            try:
                ret = cache.get(cache_key, callback=invalidate_callback)
            except KeyError:
                # Add our own `cache_context` to argument list if the wrapped function
                # has asked for one
                if self.add_cache_context:
                    kwargs["cache_context"] = _CacheContext.get_instance(
                        cache, cache_key
                    )

                ret = defer.maybeDeferred(preserve_fn(self.orig), obj, *args, **kwargs)
                ret = cache.set(cache_key, ret, callback=invalidate_callback)

            return make_deferred_yieldable(ret)

        wrapped = cast(_CachedFunction, _wrapped)

        if self.num_args == 1:
            wrapped.invalidate = lambda key: cache.invalidate(key[0])
            wrapped.prefill = lambda key, val: cache.prefill(key[0], val)
        else:
            wrapped.invalidate = cache.invalidate
            wrapped.invalidate_all = cache.invalidate_all
            wrapped.invalidate_many = cache.invalidate_many
            wrapped.prefill = cache.prefill

        wrapped.invalidate_all = cache.invalidate_all
        wrapped.cache = cache
        wrapped.num_args = self.num_args

        obj.__dict__[self.orig.__name__] = wrapped

        return wrapped


class CacheListDescriptor(_CacheDescriptorBase):
    """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 function.

    Once wrapped, the function returns a Deferred which resolves to the list
    of results.
    """

    def __init__(self, orig, cached_method_name, list_name, num_args=None):
        """
        Args:
            orig (function)
            cached_method_name (str): The name of the cached method.
            list_name (str): Name of the argument which is the bulk lookup list
            num_args (int): number of positional arguments (excluding ``self``,
                but including list_name) to use as cache keys. Defaults to all
                named args of the function.
        """
        super().__init__(orig, num_args=num_args)

        self.list_name = list_name

        self.list_pos = self.arg_names.index(self.list_name)
        self.cached_method_name = cached_method_name

        self.sentinel = object()

        if self.list_name not in self.arg_names:
            raise Exception(
                "Couldn't see arguments %r for %r."
                % (self.list_name, cached_method_name)
            )

    def __get__(self, obj, objtype=None):
        cached_method = getattr(obj, self.cached_method_name)
        cache = cached_method.cache  # type: DeferredCache[CacheKey, Any]
        num_args = cached_method.num_args

        @functools.wraps(self.orig)
        def wrapped(*args, **kwargs):
            # If we're passed a cache_context then we'll want to call its
            # invalidate() whenever we are invalidated
            invalidate_callback = kwargs.pop("on_invalidate", None)

            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]

            results = {}

            def update_results_dict(res, arg):
                results[arg] = res

            # list of deferreds to wait for
            cached_defers = []

            missing = set()

            # If the cache takes a single arg then that is used as the key,
            # otherwise a tuple is used.
            if num_args == 1:

                def arg_to_cache_key(arg):
                    return arg

            else:
                keylist = list(keyargs)

                def arg_to_cache_key(arg):
                    keylist[self.list_pos] = arg
                    return tuple(keylist)

            for arg in list_args:
                try:
                    res = cache.get(arg_to_cache_key(arg), callback=invalidate_callback)
                    if not res.called:
                        res.addCallback(update_results_dict, arg)
                        cached_defers.append(res)
                    else:
                        results[arg] = res.result
                except KeyError:
                    missing.add(arg)

            if missing:
                # we need a deferred for each entry in the list,
                # which we put in the cache. Each deferred resolves with the
                # relevant result for that key.
                deferreds_map = {}
                for arg in missing:
                    deferred = defer.Deferred()
                    deferreds_map[arg] = deferred
                    key = arg_to_cache_key(arg)
                    cache.set(key, deferred, callback=invalidate_callback)

                def complete_all(res):
                    # the wrapped function has completed. It returns a
                    # a dict. We can now resolve the observable deferreds in
                    # the cache and update our own result map.
                    for e in missing:
                        val = res.get(e, None)
                        deferreds_map[e].callback(val)
                        results[e] = val

                def errback(f):
                    # the wrapped function has failed. Invalidate any cache
                    # entries we're supposed to be populating, and fail
                    # their deferreds.
                    for e in missing:
                        key = arg_to_cache_key(e)
                        cache.invalidate(key)
                        deferreds_map[e].errback(f)

                    # return the failure, to propagate to our caller.
                    return f

                args_to_call = dict(arg_dict)
                args_to_call[self.list_name] = list(missing)

                cached_defers.append(
                    defer.maybeDeferred(
                        preserve_fn(self.orig), **args_to_call
                    ).addCallbacks(complete_all, errback)
                )

            if cached_defers:
                d = defer.gatherResults(cached_defers, consumeErrors=True).addCallbacks(
                    lambda _: results, unwrapFirstError
                )
                return make_deferred_yieldable(d)
            else:
                return defer.succeed(results)

        obj.__dict__[self.orig.__name__] = wrapped

        return wrapped


class _CacheContext:
    """Holds cache information from the cached function higher in the calling order.

    Can be used to invalidate the higher level cache entry if something changes
    on a lower level.
    """

    _cache_context_objects = (
        WeakValueDictionary()
    )  # type: WeakValueDictionary[Tuple[DeferredCache, CacheKey], _CacheContext]

    def __init__(self, cache, cache_key):  # type: (DeferredCache, CacheKey) -> None
        self._cache = cache
        self._cache_key = cache_key

    def invalidate(self):  # type: () -> None
        """Invalidates the cache entry referred to by the context."""
        self._cache.invalidate(self._cache_key)

    @classmethod
    def get_instance(
        cls, cache, cache_key
    ):  # type: (DeferredCache, CacheKey) -> _CacheContext
        """Returns an instance constructed with the given arguments.

        A new instance is only created if none already exists.
        """

        # We make sure there are no identical _CacheContext instances. This is
        # important in particular to dedupe when we add callbacks to lru cache
        # nodes, otherwise the number of callbacks would grow.
        return cls._cache_context_objects.setdefault(
            (cache, cache_key), cls(cache, cache_key)
        )


def cached(
    max_entries: int = 1000,
    num_args: Optional[int] = None,
    tree: bool = False,
    cache_context: bool = False,
    iterable: bool = False,
) -> Callable[[F], _CachedFunction[F]]:
    func = lambda orig: CacheDescriptor(
        orig,
        max_entries=max_entries,
        num_args=num_args,
        tree=tree,
        cache_context=cache_context,
        iterable=iterable,
    )

    return cast(Callable[[F], _CachedFunction[F]], func)


def cachedList(
    cached_method_name: str, list_name: str, num_args: Optional[int] = None
) -> Callable[[F], _CachedFunction[F]]:
    """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:
        cached_method_name: The name of the single-item lookup method.
            This is only used to find the cache to use.
        list_name: The name of the argument that is the list to use to
            do batch lookups in the cache.
        num_args: Number of arguments to use as the key in the cache
            (including list_name). Defaults to all named parameters.

    Example:

        class Example:
            @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):
                ...
    """
    func = lambda orig: CacheListDescriptor(
        orig,
        cached_method_name=cached_method_name,
        list_name=list_name,
        num_args=num_args,
    )

    return cast(Callable[[F], _CachedFunction[F]], func)