"""
This implements the instance caching in SQLObject. Caching is
relatively aggressive. All objects are retained so long as they are
in memory, by keeping weak references to objects. We also keep other
objects in a cache that doesn't allow them to be garbage collected
(unless caching is turned off).
"""
import threading
from weakref import ref
[docs]class CacheFactory(object):
"""
CacheFactory caches object creation. Each object should be
referenced by a single hashable ID (note tuples of hashable
values are also hashable).
"""
def __init__(self, cullFrequency=100, cullFraction=2,
cache=True):
"""
Every cullFrequency times that an item is retrieved from
this cache, the cull method is called.
The cull method then expires an arbitrary fraction of
the cached objects. The idea is at no time will the cache
be entirely emptied, placing a potentially high load at that
moment, but everything object will have its time to go
eventually. The fraction is given as an integer, and one
in that many objects are expired (i.e., the default is 1/2
of objects are expired).
By setting cache to False, items won't be cached.
However, in all cases a weak reference is kept to created
objects, and if the object hasn't been garbage collected
it will be returned.
"""
self.cullFrequency = cullFrequency
self.cullCount = 0
self.cullOffset = 0
self.cullFraction = cullFraction
self.doCache = cache
if self.doCache:
self.cache = {}
self.expiredCache = {}
self.lock = threading.Lock()
[docs] def tryGet(self, id):
"""
This returns None, or the object in cache.
"""
value = self.expiredCache.get(id)
if value:
# it's actually a weakref:
return value()
if not self.doCache:
return None
return self.cache.get(id)
[docs] def get(self, id):
"""
This method can cause deadlocks! tryGet is safer
This returns the object found in cache, or None. If None,
then the cache will remain locked! This is so that the
calling function can create the object in a threadsafe manner
before releasing the lock. You should use this like (note
that ``cache`` is actually a CacheSet object in this
example)::
obj = cache.get(some_id, my_class)
if obj is None:
try:
obj = create_object(some_id)
cache.put(some_id, my_class, obj)
finally:
cache.finishPut(cls)
This method checks both the main cache (which retains
references) and the 'expired' cache, which retains only weak
references.
"""
if self.doCache:
if self.cullCount > self.cullFrequency:
# Two threads could hit the cull in a row, but
# that's not so bad. At least by setting cullCount
# back to zero right away we avoid this. The cull
# method has a lock, so it's threadsafe.
self.cullCount = 0
self.cull()
else:
self.cullCount = self.cullCount + 1
try:
return self.cache[id]
except KeyError:
pass
self.lock.acquire()
try:
val = self.cache[id]
except KeyError:
pass
else:
self.lock.release()
return val
try:
val = self.expiredCache[id]()
except KeyError:
return None
else:
del self.expiredCache[id]
if val is None:
return None
self.cache[id] = val
self.lock.release()
return val
else:
try:
val = self.expiredCache[id]()
if val is not None:
return val
except KeyError:
pass
self.lock.acquire()
try:
val = self.expiredCache[id]()
except KeyError:
return None
else:
if val is None:
del self.expiredCache[id]
return None
self.lock.release()
return val
[docs] def put(self, id, obj):
"""
Puts an object into the cache. Should only be called after
.get(), so that duplicate objects don't end up in the cache.
"""
if self.doCache:
self.cache[id] = obj
else:
self.expiredCache[id] = ref(obj)
[docs] def finishPut(self):
"""
Releases the lock that is retained when .get() is called and
returns None.
"""
self.lock.release()
[docs] def created(self, id, obj):
"""
Inserts and object into the cache. Should be used when no one
else knows about the object yet, so there cannot be any object
already in the cache. After a database INSERT is an example
of this situation.
"""
if self.doCache:
if self.cullCount > self.cullFrequency:
# Two threads could hit the cull in a row, but
# that's not so bad. At least by setting cullCount
# back to zero right away we avoid this. The cull
# method has a lock, so it's threadsafe.
self.cullCount = 0
self.cull()
else:
self.cullCount = self.cullCount + 1
self.cache[id] = obj
else:
self.expiredCache[id] = ref(obj)
[docs] def cull(self):
"""Runs through the cache and expires objects
E.g., if ``cullFraction`` is 3, then every third object is moved to
the 'expired' (aka weakref) cache.
"""
self.lock.acquire()
try:
# remove dead references from the expired cache
keys = list(self.expiredCache.keys())
for key in keys:
if self.expiredCache[key]() is None:
self.expiredCache.pop(key, None)
keys = list(self.cache.keys())
for i in range(self.cullOffset, len(keys), self.cullFraction):
id = keys[i]
# create a weakref, then remove from the cache
obj = ref(self.cache[id])
del self.cache[id]
# the object may have been gc'd when removed from the cache
# above, no need to place in expiredCache
if obj() is not None:
self.expiredCache[id] = obj
# This offset tries to balance out which objects we
# expire, so no object will just hang out in the cache
# forever.
self.cullOffset = (self.cullOffset + 1) % self.cullFraction
finally:
self.lock.release()
[docs] def clear(self):
"""
Removes everything from the cache. Warning! This can cause
duplicate objects in memory.
"""
if self.doCache:
self.cache.clear()
self.expiredCache.clear()
[docs] def expire(self, id):
"""
Expires a single object. Typically called after a delete.
Doesn't even keep a weakref. (@@: bad name?)
"""
if not self.doCache:
return
self.lock.acquire()
try:
if id in self.cache:
del self.cache[id]
if id in self.expiredCache:
del self.expiredCache[id]
finally:
self.lock.release()
[docs] def expireAll(self):
"""
Expires all objects, moving them all into the expired/weakref
cache.
"""
if not self.doCache:
return
self.lock.acquire()
try:
for key, value in self.cache.items():
self.expiredCache[key] = ref(value)
self.cache = {}
finally:
self.lock.release()
[docs] def allIDs(self):
"""
Returns the IDs of all objects in the cache.
"""
if self.doCache:
all = list(self.cache.keys())
else:
all = []
for id, value in self.expiredCache.items():
if value():
all.append(id)
return all
[docs] def getAll(self):
"""
Return all the objects in the cache.
"""
if self.doCache:
all = list(self.cache.values())
else:
all = []
for value in self.expiredCache.values():
if value():
all.append(value())
return all
[docs]class CacheSet(object):
"""
A CacheSet is used to collect and maintain a series of caches. In
SQLObject, there is one CacheSet per connection, and one Cache
in the CacheSet for each class, since IDs are not unique across
classes. It contains methods similar to Cache, but that take
a ``cls`` argument.
"""
def __init__(self, *args, **kw):
self.caches = {}
self.args = args
self.kw = kw
[docs] def get(self, id, cls):
try:
return self.caches[cls.__name__].get(id)
except KeyError:
self.caches[cls.__name__] = CacheFactory(*self.args, **self.kw)
return self.caches[cls.__name__].get(id)
[docs] def put(self, id, cls, obj):
self.caches[cls.__name__].put(id, obj)
[docs] def finishPut(self, cls):
self.caches[cls.__name__].finishPut()
[docs] def created(self, id, cls, obj):
try:
self.caches[cls.__name__].created(id, obj)
except KeyError:
self.caches[cls.__name__] = CacheFactory(*self.args, **self.kw)
self.caches[cls.__name__].created(id, obj)
[docs] def expire(self, id, cls):
try:
self.caches[cls.__name__].expire(id)
except KeyError:
pass
[docs] def clear(self, cls=None):
if cls is None:
for cache in self.caches.values():
cache.clear()
elif cls.__name__ in self.caches:
self.caches[cls.__name__].clear()
[docs] def tryGet(self, id, cls):
return self.tryGetByName(id, cls.__name__)
[docs] def tryGetByName(self, id, clsname):
try:
return self.caches[clsname].tryGet(id)
except KeyError:
return None
[docs] def allIDs(self, cls):
try:
self.caches[cls.__name__].allIDs()
except KeyError:
return []
[docs] def allSubCaches(self):
return self.caches.values()
[docs] def allSubCachesByClassNames(self):
return self.caches
[docs] def weakrefAll(self, cls=None):
"""
Move all objects in the cls (or if not given, then in all
classes) to the weakref dictionary, where they can be
collected.
"""
if cls is None:
for cache in self.caches.values():
cache.expireAll()
elif cls.__name__ in self.caches:
self.caches[cls.__name__].expireAll()
[docs] def getAll(self, cls=None):
"""
Returns all instances in the cache for the given class or all
classes.
"""
if cls is None:
results = []
for cache in self.caches.values():
results.extend(cache.getAll())
return results
elif cls.__name__ in self.caches:
return self.caches[cls.__name__].getAll()
else:
return []