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
|
from functools import wraps
import time
try:
from collections.abc import MutableMapping # Python 3.3+
except ImportError:
from collections import MutableMapping # Python 2.7+
import heapq
from threading import Lock
class _ExpiringMapping(MutableMapping):
_INDEX = "_index_"
def __init__(self, mapping=None, capacity=None, expires_in=None, lock=None,
*args, **kwargs):
"""Items in this mapping can have individual shelf life,
just like food items in your refrigerator have their different shelf life
determined by each food, not by the refrigerator.
Expired items will be automatically evicted.
The clean-up will be done at each time when adding a new item,
or when looping or counting the entire mapping.
(This is better than being done indecisively by a background thread,
which might not always happen before your accessing the mapping.)
This implementation uses no dependency other than Python standard library.
:param MutableMapping mapping:
A dict-like key-value mapping, which needs to support __setitem__(),
__getitem__(), __delitem__(), get(), pop().
The default mapping is an in-memory dict.
You could potentially supply a file-based dict-like object, too.
This implementation deliberately avoid mapping.__iter__(),
which could be slow on a file-based mapping.
:param int capacity:
How many items this mapping will hold.
When you attempt to add new item into a full mapping,
it will automatically delete the item that is expiring soonest.
The default value is None, which means there is no capacity limit.
:param int expires_in:
How many seconds an item would expire and be purged from this mapping.
Also known as time-to-live (TTL).
You can also use :func:`~set()` to provide per-item expires_in value.
:param Lock lock:
A locking mechanism with context manager interface.
If no lock is provided, a threading.Lock will be used.
But you may want to supply a different lock,
if your customized mapping is being shared differently.
"""
super(_ExpiringMapping, self).__init__(*args, **kwargs)
self._mapping = mapping if mapping is not None else {}
self._capacity = capacity
self._expires_in = expires_in
self._lock = Lock() if lock is None else lock
def _peek(self):
# Returns (sequence, timestamps) without triggering maintenance
return self._mapping.get(self._INDEX, ([], {}))
def _validate_key(self, key):
if key == self._INDEX:
raise ValueError("key {} is a reserved keyword in {}".format(
key, self.__class__.__name__))
def set(self, key, value, expires_in):
# This method's name was chosen so that it matches its cousin __setitem__(),
# and it also complements the counterpart get().
# The downside is such a name shadows the built-in type set in this file,
# but you can overcome that by defining a global alias for set.
"""It sets the key-value pair into this mapping, with its per-item expires_in.
It will take O(logN) time, because it will run some maintenance.
This worse-than-constant time is acceptable, because in a cache scenario,
__setitem__() would only be called during a cache miss,
which would already incur an expensive target function call anyway.
By the way, most other methods of this mapping still have O(1) constant time.
"""
with self._lock:
self._set(key, value, expires_in)
def _set(self, key, value, expires_in):
# This internal implementation powers both set() and __setitem__(),
# so that they don't depend on each other.
self._validate_key(key)
sequence, timestamps = self._peek()
self._maintenance(sequence, timestamps) # O(logN)
now = int(time.time())
expires_at = now + expires_in
entry = [expires_at, now, key]
is_new_item = key not in timestamps
is_beyond_capacity = self._capacity and len(timestamps) >= self._capacity
if is_new_item and is_beyond_capacity:
self._drop_indexed_entry(timestamps, heapq.heappushpop(sequence, entry))
else: # Simply add new entry. The old one would become a harmless orphan.
heapq.heappush(sequence, entry)
timestamps[key] = [expires_at, now] # It overwrites existing key, if any
self._mapping[key] = value
self._mapping[self._INDEX] = sequence, timestamps
def _maintenance(self, sequence, timestamps): # O(logN)
"""It will modify input sequence and timestamps in-place"""
now = int(time.time())
while sequence: # Clean up expired items
expires_at, created_at, key = sequence[0]
if created_at <= now < expires_at: # Then all remaining items are fresh
break
self._drop_indexed_entry(timestamps, sequence[0]) # It could error out
heapq.heappop(sequence) # Only pop it after a successful _drop_indexed_entry()
while self._capacity is not None and len(timestamps) > self._capacity:
self._drop_indexed_entry(timestamps, sequence[0]) # It could error out
heapq.heappop(sequence) # Only pop it after a successful _drop_indexed_entry()
def _drop_indexed_entry(self, timestamps, entry):
"""For an entry came from index, drop it from timestamps and self._mapping"""
expires_at, created_at, key = entry
if [expires_at, created_at] == timestamps.get(key): # So it is not an orphan
self._mapping.pop(key, None) # It could raise exception
timestamps.pop(key, None) # This would probably always succeed
def __setitem__(self, key, value):
"""Implements the __setitem__().
Same characteristic as :func:`~set()`,
but use class-wide expires_in which was specified by :func:`~__init__()`.
"""
if self._expires_in is None:
raise ValueError("Need a numeric value for expires_in during __init__()")
with self._lock:
self._set(key, value, self._expires_in)
def __getitem__(self, key): # O(1)
"""If the item you requested already expires, KeyError will be raised."""
self._validate_key(key)
with self._lock:
# Skip self._maintenance(), because it would need O(logN) time
sequence, timestamps = self._peek()
expires_at, created_at = timestamps[key] # Would raise KeyError accordingly
now = int(time.time())
if not created_at <= now < expires_at:
self._mapping.pop(key, None)
timestamps.pop(key, None)
self._mapping[self._INDEX] = sequence, timestamps
raise KeyError("{} {}".format(
key,
"expired" if now >= expires_at else "created in the future?",
))
return self._mapping[key] # O(1)
def __delitem__(self, key): # O(1)
"""If the item you requested already expires, KeyError will be raised."""
self._validate_key(key)
with self._lock:
# Skip self._maintenance(), because it would need O(logN) time
self._mapping.pop(key, None) # O(1)
sequence, timestamps = self._peek()
del timestamps[key] # O(1)
self._mapping[self._INDEX] = sequence, timestamps
def __len__(self): # O(logN)
"""Drop all expired items and return the remaining length"""
with self._lock:
sequence, timestamps = self._peek()
self._maintenance(sequence, timestamps) # O(logN)
self._mapping[self._INDEX] = sequence, timestamps
return len(timestamps) # Faster than iter(self._mapping) when it is on disk
def __iter__(self):
"""Drop all expired items and return an iterator of the remaining items"""
with self._lock:
sequence, timestamps = self._peek()
self._maintenance(sequence, timestamps) # O(logN)
self._mapping[self._INDEX] = sequence, timestamps
return iter(timestamps) # Faster than iter(self._mapping) when it is on disk
class _IndividualCache(object):
# The code structure below can decorate both function and method.
# It is inspired by https://stackoverflow.com/a/9417088
# We may potentially switch to build upon
# https://github.com/micheles/decorator/blob/master/docs/documentation.md#statement-of-the-problem
def __init__(self, mapping=None, key_maker=None, expires_in=None):
"""Constructs a cache decorator that allows item-by-item control on
how to cache the return value of the decorated function.
:param MutableMapping mapping:
The cached items will be stored inside.
You'd want to use a ExpiringMapping
if you plan to utilize the ``expires_in`` behavior.
If nothing is provided, an in-memory dict will be used,
but it will provide no expiry functionality.
.. note::
When using this class as a decorator,
your mapping needs to be available at "compile" time,
so it would typically be a global-, module- or class-level mapping::
module_mapping = {}
@IndividualCache(mapping=module_mapping, ...)
def foo():
...
If you want to use a mapping available only at run-time,
you have to manually decorate your function at run-time, too::
def foo():
...
def bar(runtime_mapping):
foo = IndividualCache(mapping=runtime_mapping...)(foo)
:param callable key_maker:
A callable which should have signature as
``lambda function, args, kwargs: "return a string as key"``.
If key_maker happens to return ``None``, the cache will be bypassed,
the underlying function will be invoked directly,
and the invoke result will not be cached either.
:param callable expires_in:
The default value is ``None``,
which means the content being cached has no per-item expiry,
and will subject to the underlying mapping's global expiry time.
It can be an integer indicating
how many seconds the result will be cached.
In particular, if the value is 0,
it means the result expires after zero second (i.e. immediately),
therefore the result will *not* be cached.
(Mind the difference between ``expires_in=0`` and ``expires_in=None``.)
Or it can be a callable with the signature as
``lambda function=function, args=args, kwargs=kwargs, result=result: 123``
to calculate the expiry on the fly.
Its return value will be interpreted in the same way as above.
"""
self._mapping = mapping if mapping is not None else {}
self._key_maker = key_maker or (lambda function, args, kwargs: (
function, # This default implementation uses function as part of key,
# so that the cache is partitioned by function.
# However, you could have many functions to use same namespace,
# so different decorators could share same cache.
args,
tuple(kwargs.items()), # raw kwargs is not hashable
))
self._expires_in = expires_in
def __call__(self, function):
@wraps(function)
def wrapper(*args, **kwargs):
key = self._key_maker(function, args, kwargs)
if key is None: # Then bypass the cache
return function(*args, **kwargs)
now = int(time.time())
try:
return self._mapping[key]
except KeyError:
# We choose to NOT call function(...) in this block, otherwise
# potential exception from function(...) would become a confusing
# "During handling of the above exception, another exception occurred"
pass
value = function(*args, **kwargs)
expires_in = self._expires_in(
function=function,
args=args,
kwargs=kwargs,
result=value,
) if callable(self._expires_in) else self._expires_in
if expires_in == 0:
return value
if expires_in is None:
self._mapping[key] = value
else:
self._mapping.set(key, value, expires_in)
return value
return wrapper
|