#!/usr/bin/env python # -*- coding: utf-8 -*- """ | This file is part of the web2py Web Framework | Copyrighted by Massimo Di Pierro | License: LGPLv3 (http://www.gnu.org/licenses/lgpl.html) Basic caching classes and methods --------------------------------- - Cache - The generic caching object interfacing with the others - CacheInRam - providing caching in ram - CacheOnDisk - provides caches on disk Memcache is also available via a different module (see gluon.contrib.memcache) When web2py is running on Google App Engine, caching will be provided by the GAE memcache (see gluon.contrib.gae_memcache) """ import time import os import gc import sys import logging import re import random import hashlib import datetime import tempfile from gluon import recfile from collections import defaultdict from collections import OrderedDict try: from gluon import settings have_settings = True except ImportError: have_settings = False from pydal.contrib import portalocker from gluon._compat import pickle, thread, to_bytes, to_native, hashlib_md5 try: import psutil HAVE_PSUTIL = True except ImportError: HAVE_PSUTIL = False def remove_oldest_entries(storage, percentage=90): # compute current memory usage (%) old_mem = psutil.virtual_memory().percent # if we have data in storage and utilization exceeds 90% while storage and old_mem > percentage: # removed oldest entry storage.popitem(last=False) # garbage collect gc.collect(1) # comute used memory again new_mem = psutil.virtual_memory().percent # if the used memory did not decrease stop if new_mem >= old_mem: break # net new measurement for memory usage and loop old_mem = new_mem logger = logging.getLogger("web2py.cache") __all__ = ['Cache', 'lazy_cache'] DEFAULT_TIME_EXPIRE = 300 class CacheAbstract(object): """ Abstract class for cache implementations. Main function just provides referenced api documentation. Use CacheInRam or CacheOnDisk instead which are derived from this class. Note: Michele says: there are signatures inside gdbm files that are used directly by the python gdbm adapter that often are lagging behind in the detection code in python part. On every occasion that a gdbm store is probed by the python adapter, the probe fails, because gdbm file version is newer. Using gdbm directly from C would work, because there is backward compatibility, but not from python! The .shelve file is discarded and a new one created (with new signature) and it works until it is probed again... The possible consequences are memory leaks and broken sessions. """ cache_stats_name = 'web2py_cache_statistics' max_ram_utilization = None # percent def __init__(self, request=None): """Initializes the object Args: request: the global request object """ raise NotImplementedError def __call__(self, key, f, time_expire=DEFAULT_TIME_EXPIRE): """ Tries to retrieve the value corresponding to `key` from the cache if the object exists and if it did not expire, else it calls the function `f` and stores the output in the cache corresponding to `key`. It always returns the function that is returned. Args: key(str): the key of the object to be stored or retrieved f(function): the function whose output is to be cached. If `f` is `None` the cache is cleared. time_expire(int): expiration of the cache in seconds. It's used to compare the current time with the time when the requested object was last saved in cache. It does not affect future requests. Setting `time_expire` to 0 or negative value forces the cache to refresh. """ raise NotImplementedError def clear(self, regex=None): """ Clears the cache of all keys that match the provided regular expression. If no regular expression is provided, it clears all entries in cache. Args: regex: if provided, only keys matching the regex will be cleared, otherwise all keys are cleared. """ raise NotImplementedError def increment(self, key, value=1): """ Increments the cached value for the given key by the amount in value Args: key(str): key for the cached object to be incremeneted value(int): amount of the increment (defaults to 1, can be negative) """ raise NotImplementedError def _clear(self, storage, regex): """ Auxiliary function called by `clear` to search and clear cache entries """ r = re.compile(regex) for key in list(storage.keys()): if r.match(str(key)): del storage[key] return class CacheInRam(CacheAbstract): """ Ram based caching This is implemented as global (per process, shared by all threads) dictionary. A mutex-lock mechanism avoid conflicts. """ locker = thread.allocate_lock() meta_storage = {} stats = {} def __init__(self, request=None): self.initialized = False self.request = request self.storage = OrderedDict() if HAVE_PSUTIL else {} self.app = request.application if request else '' def initialize(self): if self.initialized: return else: self.initialized = True self.locker.acquire() if self.app not in self.meta_storage: self.storage = self.meta_storage[self.app] = \ OrderedDict() if HAVE_PSUTIL else {} self.stats[self.app] = {'hit_total': 0, 'misses': 0} else: self.storage = self.meta_storage[self.app] self.locker.release() def clear(self, regex=None): self.initialize() self.locker.acquire() storage = self.storage if regex is None: storage.clear() else: self._clear(storage, regex) if self.app not in self.stats: self.stats[self.app] = {'hit_total': 0, 'misses': 0} self.locker.release() def __call__(self, key, f, time_expire=DEFAULT_TIME_EXPIRE, destroyer=None): """ Attention! cache.ram does not copy the cached object. It just stores a reference to it. Turns out the deepcopying the object has some problems: - would break backward compatibility - would be limiting because people may want to cache live objects - would work unless we deepcopy no storage and retrival which would make things slow. Anyway. You can deepcopy explicitly in the function generating the value to be cached. """ self.initialize() dt = time_expire now = time.time() self.locker.acquire() item = self.storage.get(key, None) if item and f is None: del self.storage[key] if destroyer: destroyer(item[1]) self.stats[self.app]['hit_total'] += 1 self.locker.release() if f is None: return None if item and (dt is None or item[0] > now - dt): return item[1] elif item and (item[0] < now - dt) and destroyer: destroyer(item[1]) value = f() self.locker.acquire() self.storage[key] = (now, value) self.stats[self.app]['misses'] += 1 if HAVE_PSUTIL and self.max_ram_utilization is not None and random.random() < 0.10: remove_oldest_entries(self.storage, percentage=self.max_ram_utilization) self.locker.release() return value def increment(self, key, value=1): self.initialize() self.locker.acquire() try: if key in self.storage: value = self.storage[key][1] + value self.storage[key] = (time.time(), value) except BaseException as e: self.locker.release() raise e self.locker.release() return value class CacheOnDisk(CacheAbstract): """ Disk based cache This is implemented as a key value store where each key corresponds to a single file in disk which is replaced when the value changes. Disk cache provides persistance when web2py is started/stopped but it is slower than `CacheInRam` Values stored in disk cache must be pickable. """ class PersistentStorage(object): """ Implements a key based thread/process-safe safe storage in disk. """ def __init__(self, folder, file_lock_time_wait=0.1): self.folder = folder self.key_filter_in = lambda key: key self.key_filter_out = lambda key: key self.file_lock_time_wait = file_lock_time_wait # How long we should wait before retrying to lock a file held by another process # We still need a mutex for each file as portalocker only blocks other processes self.file_locks = defaultdict(thread.allocate_lock) # Make sure we use valid filenames. if sys.platform == "win32": import base64 def key_filter_in_windows(key): """ Windows doesn't allow \ / : * ? "< > | in filenames. To go around this encode the keys with base32. """ return to_native(base64.b32encode(to_bytes(key))) def key_filter_out_windows(key): """ We need to decode the keys so regex based removal works. """ return to_native(base64.b32decode(to_bytes(key))) self.key_filter_in = key_filter_in_windows self.key_filter_out = key_filter_out_windows def wait_portalock(self, val_file): """ Wait for the process file lock. """ while True: try: portalocker.lock(val_file, portalocker.LOCK_EX) break except: time.sleep(self.file_lock_time_wait) def acquire(self, key): self.file_locks[key].acquire() def release(self, key): self.file_locks[key].release() def __setitem__(self, key, value): key = self.key_filter_in(key) val_file = recfile.open(key, mode='wb', path=self.folder) self.wait_portalock(val_file) pickle.dump(value, val_file, pickle.HIGHEST_PROTOCOL) val_file.close() def __getitem__(self, key): key = self.key_filter_in(key) try: val_file = recfile.open(key, mode='rb', path=self.folder) except IOError: raise KeyError self.wait_portalock(val_file) value = pickle.load(val_file) val_file.close() return value def __contains__(self, key): key = self.key_filter_in(key) return (key in self.file_locks) or recfile.exists(key, path=self.folder) def __delitem__(self, key): key = self.key_filter_in(key) try: recfile.remove(key, path=self.folder) except IOError: raise KeyError def __iter__(self): for dirpath, dirnames, filenames in os.walk(self.folder): for filename in filenames: yield self.key_filter_out(filename) def safe_apply(self, key, function, default_value=None): """ Safely apply a function to the value of a key in storage and set the return value of the function to it. Return the result of applying the function. """ key = self.key_filter_in(key) exists = True try: val_file = recfile.open(key, mode='r+b', path=self.folder) except IOError: exists = False val_file = recfile.open(key, mode='wb', path=self.folder) self.wait_portalock(val_file) if exists: timestamp, value = pickle.load(val_file) else: value = default_value new_value = function(value) val_file.seek(0) pickle.dump((time.time(), new_value), val_file, pickle.HIGHEST_PROTOCOL) val_file.truncate() val_file.close() return new_value def keys(self): return list(self.__iter__()) def get(self, key, default=None): try: return self[key] except KeyError: return default def __init__(self, request=None, folder=None): self.initialized = False self.request = request self.folder = folder self.storage = None def initialize(self): if self.initialized: return else: self.initialized = True folder = self.folder request = self.request # Lets test if the cache folder exists, if not # we are going to create it folder = os.path.join(folder or request.folder, 'cache') if not os.path.exists(folder): os.mkdir(folder) self.storage = CacheOnDisk.PersistentStorage(folder) def __call__(self, key, f, time_expire=DEFAULT_TIME_EXPIRE): self.initialize() def inc_hit_total(v): v['hit_total'] += 1 return v def inc_misses(v): v['misses'] += 1 return v dt = time_expire self.storage.acquire(key) self.storage.acquire(CacheAbstract.cache_stats_name) item = self.storage.get(key) self.storage.safe_apply(CacheAbstract.cache_stats_name, inc_hit_total, default_value={'hit_total': 0, 'misses': 0}) if item and f is None: del self.storage[key] if f is None: self.storage.release(CacheAbstract.cache_stats_name) self.storage.release(key) return None now = time.time() if item and ((dt is None) or (item[0] > now - dt)): value = item[1] else: try: value = f() except: self.storage.release(CacheAbstract.cache_stats_name) self.storage.release(key) raise self.storage[key] = (now, value) self.storage.safe_apply(CacheAbstract.cache_stats_name, inc_misses, default_value={'hit_total': 0, 'misses': 0}) self.storage.release(CacheAbstract.cache_stats_name) self.storage.release(key) return value def clear(self, regex=None): self.initialize() storage = self.storage if regex is None: keys = storage else: r = re.compile(regex) keys = (key for key in storage if r.match(key)) for key in keys: storage.acquire(key) try: del storage[key] except KeyError: pass storage.release(key) def increment(self, key, value=1): self.initialize() self.storage.acquire(key) value = self.storage.safe_apply(key, lambda x: x + value, default_value=0) self.storage.release(key) return value class CacheAction(object): def __init__(self, func, key, time_expire, cache, cache_model): self.__name__ = func.__name__ self.__doc__ = func.__doc__ self.func = func self.key = key self.time_expire = time_expire self.cache = cache self.cache_model = cache_model def __call__(self, *a, **b): if not self.key: key2 = self.__name__ + ':' + repr(a) + ':' + repr(b) else: key2 = self.key.replace('%(name)s', self.__name__)\ .replace('%(args)s', str(a)).replace('%(vars)s', str(b)) cache_model = self.cache_model if not cache_model or isinstance(cache_model, str): cache_model = getattr(self.cache, cache_model or 'ram') return cache_model(key2, lambda a=a, b=b: self.func(*a, **b), self.time_expire) class Cache(object): """ Sets up generic caching, creating an instance of both CacheInRam and CacheOnDisk. In case of GAE will make use of gluon.contrib.gae_memcache. - self.ram is an instance of CacheInRam - self.disk is an instance of CacheOnDisk """ autokey = ':%(name)s:%(args)s:%(vars)s' def __init__(self, request): """ Args: request: the global request object """ # GAE will have a special caching if have_settings and settings.global_settings.web2py_runtime_gae: from gluon.contrib.gae_memcache import MemcacheClient self.ram = self.disk = MemcacheClient(request) else: # Otherwise use ram (and try also disk) self.ram = CacheInRam(request) try: self.disk = CacheOnDisk(request) except IOError: logger.warning('no cache.disk (IOError)') except AttributeError: # normally not expected anymore, as GAE has already # been accounted for logger.warning('no cache.disk (AttributeError)') def action(self, time_expire=DEFAULT_TIME_EXPIRE, cache_model=None, prefix=None, session=False, vars=True, lang=True, user_agent=False, public=True, valid_statuses=None, quick=None): """Better fit for caching an action Warning: Experimental! Currently only HTTP 1.1 compliant reference : http://code.google.com/p/doctype-mirror/wiki/ArticleHttpCaching Args: time_expire(int): same as @cache cache_model(str): same as @cache prefix(str): add a prefix to the calculated key session(bool): adds response.session_id to the key vars(bool): adds request.env.query_string lang(bool): adds T.accepted_language user_agent(bool or dict): if True, adds is_mobile and is_tablet to the key. Pass a dict to use all the needed values (uses str(.items())) (e.g. user_agent=request.user_agent()). Used only if session is not True public(bool): if False forces the Cache-Control to be 'private' valid_statuses: by default only status codes starting with 1,2,3 will be cached. pass an explicit list of statuses on which turn the cache on quick: Session,Vars,Lang,User-agent,Public: fast overrides with initials, e.g. 'SVLP' or 'VLP', or 'VLP' """ from gluon import current from gluon.http import HTTP def wrap(func): def wrapped_f(): if current.request.env.request_method != 'GET': return func() if quick: session_ = True if 'S' in quick else False vars_ = True if 'V' in quick else False lang_ = True if 'L' in quick else False user_agent_ = True if 'U' in quick else False public_ = True if 'P' in quick else False else: (session_, vars_, lang_, user_agent_, public_) = \ (session, vars, lang, user_agent, public) if time_expire: cache_control = 'max-age=%(time_expire)s, s-maxage=%(time_expire)s' % dict(time_expire=time_expire) if not session_ and public_: cache_control += ', public' expires = (current.request.utcnow + datetime.timedelta(seconds=time_expire) ).strftime('%a, %d %b %Y %H:%M:%S GMT') else: cache_control += ', private' expires = 'Fri, 01 Jan 1990 00:00:00 GMT' if cache_model: # figure out the correct cache key cache_key = [current.request.env.path_info, current.response.view] if session_: cache_key.append(current.response.session_id) elif user_agent_: if user_agent_ is True: cache_key.append("%(is_mobile)s_%(is_tablet)s" % current.request.user_agent()) else: cache_key.append(str(user_agent_.items())) if vars_: cache_key.append(current.request.env.query_string) if lang_: cache_key.append(current.T.accepted_language) cache_key = hashlib_md5('__'.join(cache_key)).hexdigest() if prefix: cache_key = prefix + cache_key try: # action returns something rtn = cache_model(cache_key, lambda: func(), time_expire=time_expire) http, status = None, current.response.status except HTTP as e: # action raises HTTP (can still be valid) rtn = cache_model(cache_key, lambda: e.body, time_expire=time_expire) http, status = HTTP(e.status, rtn, **e.headers), e.status else: # action raised a generic exception http = None else: # no server-cache side involved try: # action returns something rtn = func() http, status = None, current.response.status except HTTP as e: # action raises HTTP (can still be valid) status = e.status http = HTTP(e.status, e.body, **e.headers) else: # action raised a generic exception http = None send_headers = False if http and isinstance(valid_statuses, list): if status in valid_statuses: send_headers = True elif valid_statuses is None: if str(status)[0] in '123': send_headers = True if send_headers: headers = {'Pragma': None, 'Expires': expires, 'Cache-Control': cache_control} current.response.headers.update(headers) if cache_model and not send_headers: # we cached already the value, but the status is not valid # so we need to delete the cached value cache_model(cache_key, None) if http: if send_headers: http.headers.update(current.response.headers) raise http return rtn wrapped_f.__name__ = func.__name__ wrapped_f.__doc__ = func.__doc__ return wrapped_f return wrap def __call__(self, key=None, time_expire=DEFAULT_TIME_EXPIRE, cache_model=None): """ Decorator function that can be used to cache any function/method. Args: key(str) : the key of the object to be store or retrieved time_expire(int) : expiration of the cache in seconds `time_expire` is used to compare the current time with the time when the requested object was last saved in cache. It does not affect future requests. Setting `time_expire` to 0 or negative value forces the cache to refresh. cache_model(str): can be "ram", "disk" or other (like "memcache"). Defaults to "ram" When the function `f` is called, web2py tries to retrieve the value corresponding to `key` from the cache if the object exists and if it did not expire, else it calles the function `f` and stores the output in the cache corresponding to `key`. In the case the output of the function is returned. Example: :: @cache('key', 5000, cache.ram) def f(): return time.ctime() Note: If the function `f` is an action, we suggest using @cache.action instead """ def tmp(func, cache=self, cache_model=cache_model): return CacheAction(func, key, time_expire, self, cache_model) return tmp @staticmethod def with_prefix(cache_model, prefix): """ allow replacing cache.ram with cache.with_prefix(cache.ram,'prefix') it will add prefix to all the cache keys used. """ return lambda key, f, time_expire=DEFAULT_TIME_EXPIRE, prefix=prefix: cache_model(prefix + key, f, time_expire) def lazy_cache(key=None, time_expire=None, cache_model='ram'): """ Can be used to cache any function including ones in modules, as long as the cached function is only called within a web2py request If a key is not provided, one is generated from the function name `time_expire` defaults to None (no cache expiration) If cache_model is "ram" then the model is current.cache.ram, etc. """ def decorator(f, key=key, time_expire=time_expire, cache_model=cache_model): key = key or repr(f) def g(*c, **d): from gluon import current return current.cache(key, time_expire, cache_model)(f)(*c, **d) g.__name__ = f.__name__ return g return decorator