def test_pop(self):
m = Mock()
cache = LruCache(1)
cache.set("key", "value", callbacks=[m])
self.assertFalse(m.called)
cache.pop("key")
self.assertEquals(m.call_count, 1)
cache.set("key", "value")
self.assertEquals(m.call_count, 1)
cache.pop("key")
self.assertEquals(m.call_count, 1)
class Cache(object):
__slots__ = (
"cache",
"name",
"keylen",
"thread",
"metrics",
"_pending_deferred_cache",
)
def __init__(
self,
name: str,
max_entries: int = 1000,
keylen: int = 1,
tree: bool = False,
iterable: bool = False,
apply_cache_factor_from_config: bool = True,
):
"""
Args:
name: The name of the cache
max_entries: Maximum amount of entries that the cache will hold
keylen: The length of the tuple used as the cache key
tree: Use a TreeCache instead of a dict as the underlying cache type
iterable: If True, count each item in the cached object as an entry,
rather than each cached object
apply_cache_factor_from_config: Whether cache factors specified in the
config file affect `max_entries`
Returns:
Cache
"""
cache_type = TreeCache if tree else dict
self._pending_deferred_cache = cache_type()
self.cache = LruCache(
max_size=max_entries,
keylen=keylen,
cache_type=cache_type,
size_callback=(lambda d: len(d)) if iterable else None,
evicted_callback=self._on_evicted,
apply_cache_factor_from_config=apply_cache_factor_from_config,
)
self.name = name
self.keylen = keylen
self.thread = None
self.metrics = register_cache(
"cache",
name,
self.cache,
collect_callback=self._metrics_collection_callback,
)
@property
def max_entries(self):
return self.cache.max_size
def _on_evicted(self, evicted_count):
self.metrics.inc_evictions(evicted_count)
def _metrics_collection_callback(self):
cache_pending_metric.labels(self.name).set(
len(self._pending_deferred_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,
callback=None,
update_metrics=True):
"""Looks the key up in the caches.
Args:
key(tuple)
default: What is returned if key is not in the caches. If not
specified then function throws KeyError instead
callback(fn): Gets called when the entry in the cache is invalidated
update_metrics (bool): whether to update the cache hit rate metrics
Returns:
Either an ObservableDeferred or the raw result
"""
callbacks = [callback] if callback else []
val = self._pending_deferred_cache.get(key, _CacheSentinel)
if val is not _CacheSentinel:
val.callbacks.update(callbacks)
if update_metrics:
self.metrics.inc_hits()
return val.deferred
val = self.cache.get(key, _CacheSentinel, callbacks=callbacks)
if val is not _CacheSentinel:
self.metrics.inc_hits()
return val
if update_metrics:
self.metrics.inc_misses()
if default is _CacheSentinel:
raise KeyError()
else:
return default
def set(self, key, value, callback=None):
if not isinstance(value, defer.Deferred):
raise TypeError("not a Deferred")
callbacks = [callback] if callback else []
self.check_thread()
observable = ObservableDeferred(value, consumeErrors=True)
observer = defer.maybeDeferred(observable.observe)
entry = CacheEntry(deferred=observable, callbacks=callbacks)
existing_entry = self._pending_deferred_cache.pop(key, None)
if existing_entry:
existing_entry.invalidate()
self._pending_deferred_cache[key] = entry
def compare_and_pop():
"""Check if our entry is still the one in _pending_deferred_cache, and
if so, pop it.
Returns true if the entries matched.
"""
existing_entry = self._pending_deferred_cache.pop(key, None)
if existing_entry is entry:
return True
# oops, the _pending_deferred_cache has been updated since
# we started our query, so we are out of date.
#
# Better put back whatever we took out. (We do it this way
# round, rather than peeking into the _pending_deferred_cache
# and then removing on a match, to make the common case faster)
if existing_entry is not None:
self._pending_deferred_cache[key] = existing_entry
return False
def cb(result):
if compare_and_pop():
self.cache.set(key, result, entry.callbacks)
else:
# we're not going to put this entry into the cache, so need
# to make sure that the invalidation callbacks are called.
# That was probably done when _pending_deferred_cache was
# updated, but it's possible that `set` was called without
# `invalidate` being previously called, in which case it may
# not have been. Either way, let's double-check now.
entry.invalidate()
def eb(_fail):
compare_and_pop()
entry.invalidate()
# once the deferred completes, we can move the entry from the
# _pending_deferred_cache to the real cache.
#
observer.addCallbacks(cb, eb)
return observable
def prefill(self, key, value, callback=None):
callbacks = [callback] if callback else []
self.cache.set(key, value, callbacks=callbacks)
def invalidate(self, key):
self.check_thread()
self.cache.pop(key, None)
# if we have a pending lookup for this key, remove it from the
# _pending_deferred_cache, which will (a) stop it being returned
# for future queries and (b) stop it being persisted as a proper entry
# in self.cache.
entry = self._pending_deferred_cache.pop(key, None)
# run the invalidation callbacks now, rather than waiting for the
# deferred to resolve.
if entry:
entry.invalidate()
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.cache.del_multi(key)
# if we have a pending lookup for this key, remove it from the
# _pending_deferred_cache, as above
entry_dict = self._pending_deferred_cache.pop(key, None)
if entry_dict is not None:
for entry in iterate_tree_cache_entry(entry_dict):
entry.invalidate()
def invalidate_all(self):
self.check_thread()
self.cache.clear()
for entry in self._pending_deferred_cache.values():
entry.invalidate()
self._pending_deferred_cache.clear()
def test_pop(self):
cache = LruCache(1)
cache["key"] = 1
self.assertEquals(cache.pop("key"), 1)
self.assertEquals(cache.pop("key"), None)
class Cache(object):
__slots__ = (
"cache",
"max_entries",
"name",
"keylen",
"thread",
"metrics",
"_pending_deferred_cache",
)
def __init__(self,
name,
max_entries=1000,
keylen=1,
tree=False,
iterable=False):
cache_type = TreeCache if tree else dict
self._pending_deferred_cache = cache_type()
self.cache = LruCache(
max_size=max_entries,
keylen=keylen,
cache_type=cache_type,
size_callback=(lambda d: len(d)) if iterable else None,
evicted_callback=self._on_evicted,
)
self.name = name
self.keylen = keylen
self.thread = None
self.metrics = register_cache("cache", name, self.cache)
def _on_evicted(self, evicted_count):
self.metrics.inc_evictions(evicted_count)
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,
callback=None,
update_metrics=True):
"""Looks the key up in the caches.
Args:
key(tuple)
default: What is returned if key is not in the caches. If not
specified then function throws KeyError instead
callback(fn): Gets called when the entry in the cache is invalidated
update_metrics (bool): whether to update the cache hit rate metrics
Returns:
Either a Deferred or the raw result
"""
callbacks = [callback] if callback else []
val = self._pending_deferred_cache.get(key, _CacheSentinel)
if val is not _CacheSentinel:
val.callbacks.update(callbacks)
if update_metrics:
self.metrics.inc_hits()
return val.deferred
val = self.cache.get(key, _CacheSentinel, callbacks=callbacks)
if val is not _CacheSentinel:
self.metrics.inc_hits()
return val
if update_metrics:
self.metrics.inc_misses()
if default is _CacheSentinel:
raise KeyError()
else:
return default
def set(self, key, value, callback=None):
callbacks = [callback] if callback else []
self.check_thread()
entry = CacheEntry(
deferred=value,
callbacks=callbacks,
)
existing_entry = self._pending_deferred_cache.pop(key, None)
if existing_entry:
existing_entry.invalidate()
self._pending_deferred_cache[key] = entry
def shuffle(result):
existing_entry = self._pending_deferred_cache.pop(key, None)
if existing_entry is entry:
self.cache.set(key, result, entry.callbacks)
else:
# oops, the _pending_deferred_cache has been updated since
# we started our query, so we are out of date.
#
# Better put back whatever we took out. (We do it this way
# round, rather than peeking into the _pending_deferred_cache
# and then removing on a match, to make the common case faster)
if existing_entry is not None:
self._pending_deferred_cache[key] = existing_entry
# we're not going to put this entry into the cache, so need
# to make sure that the invalidation callbacks are called.
# That was probably done when _pending_deferred_cache was
# updated, but it's possible that `set` was called without
# `invalidate` being previously called, in which case it may
# not have been. Either way, let's double-check now.
entry.invalidate()
return result
entry.deferred.addCallback(shuffle)
def prefill(self, key, value, callback=None):
callbacks = [callback] if callback else []
self.cache.set(key, value, callbacks=callbacks)
def invalidate(self, key):
self.check_thread()
self.cache.pop(key, None)
# if we have a pending lookup for this key, remove it from the
# _pending_deferred_cache, which will (a) stop it being returned
# for future queries and (b) stop it being persisted as a proper entry
# in self.cache.
entry = self._pending_deferred_cache.pop(key, None)
# run the invalidation callbacks now, rather than waiting for the
# deferred to resolve.
if entry:
entry.invalidate()
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.cache.del_multi(key)
# if we have a pending lookup for this key, remove it from the
# _pending_deferred_cache, as above
entry_dict = self._pending_deferred_cache.pop(key, None)
if entry_dict is not None:
for entry in iterate_tree_cache_entry(entry_dict):
entry.invalidate()
def invalidate_all(self):
self.check_thread()
self.cache.clear()
for entry in itervalues(self._pending_deferred_cache):
entry.invalidate()
self._pending_deferred_cache.clear()
class DictionaryCache(object):
"""Caches key -> dictionary lookups, supporting caching partial dicts, i.e.
fetching a subset of dictionary keys for a particular key.
"""
def __init__(self, name, max_entries=1000):
self.cache = LruCache(max_size=max_entries)
self.name = name
self.sequence = 0
self.thread = None
# caches_by_name[name] = self.cache
class Sentinel(object):
__slots__ = []
self.sentinel = Sentinel()
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, dict_keys=None):
entry = self.cache.get(key, self.sentinel)
if entry is not self.sentinel:
cache_counter.inc_hits(self.name)
if dict_keys is None:
return DictionaryEntry(entry.full, dict(entry.value))
else:
return DictionaryEntry(
entry.full,
{k: entry.value[k]
for k in dict_keys if k in entry.value})
cache_counter.inc_misses(self.name)
return DictionaryEntry(False, {})
def invalidate(self, key):
self.check_thread()
# 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)
def invalidate_all(self):
self.check_thread()
self.sequence += 1
self.cache.clear()
def update(self, sequence, key, value, full=False):
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)
if full:
self._insert(key, value)
else:
self._update_or_insert(key, value)
def _update_or_insert(self, key, value):
entry = self.cache.setdefault(key, DictionaryEntry(False, {}))
entry.value.update(value)
def _insert(self, key, value):
self.cache[key] = DictionaryEntry(True, value)
class Cache(object):
__slots__ = (
"cache",
"max_entries",
"name",
"keylen",
"thread",
"metrics",
"_pending_deferred_cache",
)
def __init__(self, name, max_entries=1000, keylen=1, tree=False, iterable=False):
cache_type = TreeCache if tree else dict
self._pending_deferred_cache = cache_type()
self.cache = LruCache(
max_size=max_entries, keylen=keylen, cache_type=cache_type,
size_callback=(lambda d: len(d)) if iterable else None,
evicted_callback=self._on_evicted,
)
self.name = name
self.keylen = keylen
self.thread = None
self.metrics = register_cache("cache", name, self.cache)
def _on_evicted(self, evicted_count):
self.metrics.inc_evictions(evicted_count)
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, callback=None, update_metrics=True):
"""Looks the key up in the caches.
Args:
key(tuple)
default: What is returned if key is not in the caches. If not
specified then function throws KeyError instead
callback(fn): Gets called when the entry in the cache is invalidated
update_metrics (bool): whether to update the cache hit rate metrics
Returns:
Either a Deferred or the raw result
"""
callbacks = [callback] if callback else []
val = self._pending_deferred_cache.get(key, _CacheSentinel)
if val is not _CacheSentinel:
val.callbacks.update(callbacks)
if update_metrics:
self.metrics.inc_hits()
return val.deferred
val = self.cache.get(key, _CacheSentinel, callbacks=callbacks)
if val is not _CacheSentinel:
self.metrics.inc_hits()
return val
if update_metrics:
self.metrics.inc_misses()
if default is _CacheSentinel:
raise KeyError()
else:
return default
def set(self, key, value, callback=None):
callbacks = [callback] if callback else []
self.check_thread()
entry = CacheEntry(
deferred=value,
callbacks=callbacks,
)
existing_entry = self._pending_deferred_cache.pop(key, None)
if existing_entry:
existing_entry.invalidate()
self._pending_deferred_cache[key] = entry
def shuffle(result):
existing_entry = self._pending_deferred_cache.pop(key, None)
if existing_entry is entry:
self.cache.set(key, result, entry.callbacks)
else:
# oops, the _pending_deferred_cache has been updated since
# we started our query, so we are out of date.
#
# Better put back whatever we took out. (We do it this way
# round, rather than peeking into the _pending_deferred_cache
# and then removing on a match, to make the common case faster)
if existing_entry is not None:
self._pending_deferred_cache[key] = existing_entry
# we're not going to put this entry into the cache, so need
# to make sure that the invalidation callbacks are called.
# That was probably done when _pending_deferred_cache was
# updated, but it's possible that `set` was called without
# `invalidate` being previously called, in which case it may
# not have been. Either way, let's double-check now.
entry.invalidate()
return result
entry.deferred.addCallback(shuffle)
def prefill(self, key, value, callback=None):
callbacks = [callback] if callback else []
self.cache.set(key, value, callbacks=callbacks)
def invalidate(self, key):
self.check_thread()
self.cache.pop(key, None)
# if we have a pending lookup for this key, remove it from the
# _pending_deferred_cache, which will (a) stop it being returned
# for future queries and (b) stop it being persisted as a proper entry
# in self.cache.
entry = self._pending_deferred_cache.pop(key, None)
# run the invalidation callbacks now, rather than waiting for the
# deferred to resolve.
if entry:
entry.invalidate()
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.cache.del_multi(key)
# if we have a pending lookup for this key, remove it from the
# _pending_deferred_cache, as above
entry_dict = self._pending_deferred_cache.pop(key, None)
if entry_dict is not None:
for entry in iterate_tree_cache_entry(entry_dict):
entry.invalidate()
def invalidate_all(self):
self.check_thread()
self.cache.clear()
for entry in itervalues(self._pending_deferred_cache):
entry.invalidate()
self._pending_deferred_cache.clear()
class DeferredCache(Generic[KT, VT]):
"""Wraps an LruCache, adding support for Deferred results.
It expects that each entry added with set() will be a Deferred; likewise get()
will return a Deferred.
"""
__slots__ = (
"cache",
"thread",
"_pending_deferred_cache",
)
def __init__(
self,
name: str,
max_entries: int = 1000,
tree: bool = False,
iterable: bool = False,
apply_cache_factor_from_config: bool = True,
):
"""
Args:
name: The name of the cache
max_entries: Maximum amount of entries that the cache will hold
keylen: The length of the tuple used as the cache key. Ignored unless
`tree` is True.
tree: Use a TreeCache instead of a dict as the underlying cache type
iterable: If True, count each item in the cached object as an entry,
rather than each cached object
apply_cache_factor_from_config: Whether cache factors specified in the
config file affect `max_entries`
"""
cache_type = TreeCache if tree else dict
# _pending_deferred_cache maps from the key value to a `CacheEntry` object.
self._pending_deferred_cache = (
cache_type()) # type: MutableMapping[KT, CacheEntry]
def metrics_cb():
cache_pending_metric.labels(name).set(
len(self._pending_deferred_cache))
# cache is used for completed results and maps to the result itself, rather than
# a Deferred.
self.cache = LruCache(
max_size=max_entries,
cache_name=name,
cache_type=cache_type,
size_callback=(lambda d: len(d) or 1) if iterable else None,
metrics_collection_callback=metrics_cb,
apply_cache_factor_from_config=apply_cache_factor_from_config,
) # type: LruCache[KT, VT]
self.thread = None # type: Optional[threading.Thread]
@property
def max_entries(self):
return self.cache.max_size
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: KT,
callback: Optional[Callable[[], None]] = None,
update_metrics: bool = True,
) -> defer.Deferred:
"""Looks the key up in the caches.
For symmetry with set(), this method does *not* follow the synapse logcontext
rules: the logcontext will not be cleared on return, and the Deferred will run
its callbacks in the sentinel context. In other words: wrap the result with
make_deferred_yieldable() before `await`ing it.
Args:
key:
callback: Gets called when the entry in the cache is invalidated
update_metrics (bool): whether to update the cache hit rate metrics
Returns:
A Deferred which completes with the result. Note that this may later fail
if there is an ongoing set() operation which later completes with a failure.
Raises:
KeyError if the key is not found in the cache
"""
callbacks = [callback] if callback else []
val = self._pending_deferred_cache.get(key, _Sentinel.sentinel)
if val is not _Sentinel.sentinel:
val.callbacks.update(callbacks)
if update_metrics:
m = self.cache.metrics
assert m # we always have a name, so should always have metrics
m.inc_hits()
return val.deferred.observe()
val2 = self.cache.get(key,
_Sentinel.sentinel,
callbacks=callbacks,
update_metrics=update_metrics)
if val2 is _Sentinel.sentinel:
raise KeyError()
else:
return defer.succeed(val2)
def get_immediate(self,
key: KT,
default: T,
update_metrics: bool = True) -> Union[VT, T]:
"""If we have a *completed* cached value, return it."""
return self.cache.get(key, default, update_metrics=update_metrics)
def set(
self,
key: KT,
value: defer.Deferred,
callback: Optional[Callable[[], None]] = None,
) -> defer.Deferred:
"""Adds a new entry to the cache (or updates an existing one).
The given `value` *must* be a Deferred.
First any existing entry for the same key is invalidated. Then a new entry
is added to the cache for the given key.
Until the `value` completes, calls to `get()` for the key will also result in an
incomplete Deferred, which will ultimately complete with the same result as
`value`.
If `value` completes successfully, subsequent calls to `get()` will then return
a completed deferred with the same result. If it *fails*, the cache is
invalidated and subequent calls to `get()` will raise a KeyError.
If another call to `set()` happens before `value` completes, then (a) any
invalidation callbacks registered in the interim will be called, (b) any
`get()`s in the interim will continue to complete with the result from the
*original* `value`, (c) any future calls to `get()` will complete with the
result from the *new* `value`.
It is expected that `value` does *not* follow the synapse logcontext rules - ie,
if it is incomplete, it runs its callbacks in the sentinel context.
Args:
key: Key to be set
value: a deferred which will complete with a result to add to the cache
callback: An optional callback to be called when the entry is invalidated
"""
if not isinstance(value, defer.Deferred):
raise TypeError("not a Deferred")
callbacks = [callback] if callback else []
self.check_thread()
existing_entry = self._pending_deferred_cache.pop(key, None)
if existing_entry:
existing_entry.invalidate()
# XXX: why don't we invalidate the entry in `self.cache` yet?
# we can save a whole load of effort if the deferred is ready.
if value.called:
result = value.result
if not isinstance(result, failure.Failure):
self.cache.set(key, result, callbacks)
return value
# otherwise, we'll add an entry to the _pending_deferred_cache for now,
# and add callbacks to add it to the cache properly later.
observable = ObservableDeferred(value, consumeErrors=True)
observer = observable.observe()
entry = CacheEntry(deferred=observable, callbacks=callbacks)
self._pending_deferred_cache[key] = entry
def compare_and_pop():
"""Check if our entry is still the one in _pending_deferred_cache, and
if so, pop it.
Returns true if the entries matched.
"""
existing_entry = self._pending_deferred_cache.pop(key, None)
if existing_entry is entry:
return True
# oops, the _pending_deferred_cache has been updated since
# we started our query, so we are out of date.
#
# Better put back whatever we took out. (We do it this way
# round, rather than peeking into the _pending_deferred_cache
# and then removing on a match, to make the common case faster)
if existing_entry is not None:
self._pending_deferred_cache[key] = existing_entry
return False
def cb(result):
if compare_and_pop():
self.cache.set(key, result, entry.callbacks)
else:
# we're not going to put this entry into the cache, so need
# to make sure that the invalidation callbacks are called.
# That was probably done when _pending_deferred_cache was
# updated, but it's possible that `set` was called without
# `invalidate` being previously called, in which case it may
# not have been. Either way, let's double-check now.
entry.invalidate()
def eb(_fail):
compare_and_pop()
entry.invalidate()
# once the deferred completes, we can move the entry from the
# _pending_deferred_cache to the real cache.
#
observer.addCallbacks(cb, eb)
# we return a new Deferred which will be called before any subsequent observers.
return observable.observe()
def prefill(self,
key: KT,
value: VT,
callback: Optional[Callable[[], None]] = None):
callbacks = [callback] if callback else []
self.cache.set(key, value, callbacks=callbacks)
def invalidate(self, key):
self.check_thread()
self.cache.pop(key, None)
# if we have a pending lookup for this key, remove it from the
# _pending_deferred_cache, which will (a) stop it being returned
# for future queries and (b) stop it being persisted as a proper entry
# in self.cache.
entry = self._pending_deferred_cache.pop(key, None)
# run the invalidation callbacks now, rather than waiting for the
# deferred to resolve.
if entry:
entry.invalidate()
def invalidate_many(self, key: KT):
self.check_thread()
if not isinstance(key, tuple):
raise TypeError("The cache key must be a tuple not %r" %
(type(key), ))
key = cast(KT, key)
self.cache.del_multi(key)
# if we have a pending lookup for this key, remove it from the
# _pending_deferred_cache, as above
entry_dict = self._pending_deferred_cache.pop(key, None)
if entry_dict is not None:
for entry in iterate_tree_cache_entry(entry_dict):
entry.invalidate()
def invalidate_all(self):
self.check_thread()
self.cache.clear()
for entry in self._pending_deferred_cache.values():
entry.invalidate()
self._pending_deferred_cache.clear()
class Cache(object):
__slots__ = (
"cache",
"max_entries",
"name",
"keylen",
"sequence",
"thread",
"metrics",
"_pending_deferred_cache",
)
def __init__(self, name, max_entries=1000, keylen=1, tree=False, iterable=False):
cache_type = TreeCache if tree else dict
self._pending_deferred_cache = cache_type()
self.cache = LruCache(
max_size=max_entries, keylen=keylen, cache_type=cache_type,
size_callback=(lambda d: len(d)) if iterable else None,
)
self.name = name
self.keylen = keylen
self.sequence = 0
self.thread = None
self.metrics = register_cache(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, callback=None, update_metrics=True):
"""Looks the key up in the caches.
Args:
key(tuple)
default: What is returned if key is not in the caches. If not
specified then function throws KeyError instead
callback(fn): Gets called when the entry in the cache is invalidated
update_metrics (bool): whether to update the cache hit rate metrics
Returns:
Either a Deferred or the raw result
"""
callbacks = [callback] if callback else []
val = self._pending_deferred_cache.get(key, _CacheSentinel)
if val is not _CacheSentinel:
if val.sequence == self.sequence:
val.callbacks.update(callbacks)
if update_metrics:
self.metrics.inc_hits()
return val.deferred
val = self.cache.get(key, _CacheSentinel, callbacks=callbacks)
if val is not _CacheSentinel:
self.metrics.inc_hits()
return val
if update_metrics:
self.metrics.inc_misses()
if default is _CacheSentinel:
raise KeyError()
else:
return default
def set(self, key, value, callback=None):
callbacks = [callback] if callback else []
self.check_thread()
entry = CacheEntry(
deferred=value,
sequence=self.sequence,
callbacks=callbacks,
)
entry.callbacks.update(callbacks)
existing_entry = self._pending_deferred_cache.pop(key, None)
if existing_entry:
existing_entry.invalidate()
self._pending_deferred_cache[key] = entry
def shuffle(result):
if self.sequence == entry.sequence:
existing_entry = self._pending_deferred_cache.pop(key, None)
if existing_entry is entry:
self.cache.set(key, result, entry.callbacks)
else:
entry.invalidate()
else:
entry.invalidate()
return result
entry.deferred.addCallback(shuffle)
def prefill(self, key, value, callback=None):
callbacks = [callback] if callback else []
self.cache.set(key, value, callbacks=callbacks)
def invalidate(self, key):
self.check_thread()
# Increment the sequence number so that any SELECT statements that
# raced with the INSERT don't update the cache (SYN-369)
self.sequence += 1
entry = self._pending_deferred_cache.pop(key, None)
if entry:
entry.invalidate()
self.cache.pop(key, None)
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)
entry_dict = self._pending_deferred_cache.pop(key, None)
if entry_dict is not None:
for entry in iterate_tree_cache_entry(entry_dict):
entry.invalidate()
def invalidate_all(self):
self.check_thread()
self.sequence += 1
self.cache.clear()
class DictionaryCache:
"""Caches key -> dictionary lookups, supporting caching partial dicts, i.e.
fetching a subset of dictionary keys for a particular key.
"""
def __init__(self, name, max_entries=1000):
self.cache = LruCache(
max_size=max_entries, cache_name=name,
size_callback=len) # type: LruCache[Any, DictionaryEntry]
self.name = name
self.sequence = 0
self.thread = None
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, dict_keys=None):
"""Fetch an entry out of the cache
Args:
key
dict_key(list): If given a set of keys then return only those keys
that exist in the cache.
Returns:
DictionaryEntry
"""
entry = self.cache.get(key, _Sentinel.sentinel)
if entry is not _Sentinel.sentinel:
if dict_keys is None:
return DictionaryEntry(entry.full, entry.known_absent,
dict(entry.value))
else:
return DictionaryEntry(
entry.full,
entry.known_absent,
{k: entry.value[k]
for k in dict_keys if k in entry.value},
)
return DictionaryEntry(False, set(), {})
def invalidate(self, key):
self.check_thread()
# 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)
def invalidate_all(self):
self.check_thread()
self.sequence += 1
self.cache.clear()
def update(self, sequence, key, value, fetched_keys=None):
"""Updates the entry in the cache
Args:
sequence
key (K)
value (dict[X,Y]): The value to update the cache with.
fetched_keys (None|set[X]): All of the dictionary keys which were
fetched from the database.
If None, this is the complete value for key K. Otherwise, it
is used to infer a list of keys which we know don't exist in
the full dict.
"""
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)
if fetched_keys is None:
self._insert(key, value, set())
else:
self._update_or_insert(key, value, fetched_keys)
def _update_or_insert(self, key, value, known_absent):
# We pop and reinsert as we need to tell the cache the size may have
# changed
entry = self.cache.pop(key, DictionaryEntry(False, set(), {}))
entry.value.update(value)
entry.known_absent.update(known_absent)
self.cache[key] = entry
def _insert(self, key, value, known_absent):
self.cache[key] = DictionaryEntry(True, known_absent, value)
class DictionaryCache(object):
"""Caches key -> dictionary lookups, supporting caching partial dicts, i.e.
fetching a subset of dictionary keys for a particular key.
"""
def __init__(self, name, max_entries=1000):
self.cache = LruCache(max_size=max_entries, size_callback=len)
self.name = name
self.sequence = 0
self.thread = None
# caches_by_name[name] = self.cache
class Sentinel(object):
__slots__ = []
self.sentinel = Sentinel()
self.metrics = register_cache(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, dict_keys=None):
"""Fetch an entry out of the cache
Args:
key
dict_key(list): If given a set of keys then return only those keys
that exist in the cache.
Returns:
DictionaryEntry
"""
entry = self.cache.get(key, self.sentinel)
if entry is not self.sentinel:
self.metrics.inc_hits()
if dict_keys is None:
return DictionaryEntry(entry.full, entry.known_absent,
dict(entry.value))
else:
return DictionaryEntry(
entry.full, entry.known_absent,
{k: entry.value[k]
for k in dict_keys if k in entry.value})
self.metrics.inc_misses()
return DictionaryEntry(False, set(), {})
def invalidate(self, key):
self.check_thread()
# 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)
def invalidate_all(self):
self.check_thread()
self.sequence += 1
self.cache.clear()
def update(self, sequence, key, value, full=False, known_absent=None):
"""Updates the entry in the cache
Args:
sequence
key
value (dict): The value to update the cache with.
full (bool): Whether the given value is the full dict, or just a
partial subset there of. If not full then any existing entries
for the key will be updated.
known_absent (set): Set of keys that we know don't exist in the full
dict.
"""
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)
if known_absent is None:
known_absent = set()
if full:
self._insert(key, value, known_absent)
else:
self._update_or_insert(key, value, known_absent)
def _update_or_insert(self, key, value, known_absent):
# We pop and reinsert as we need to tell the cache the size may have
# changed
entry = self.cache.pop(key, DictionaryEntry(False, set(), {}))
entry.value.update(value)
entry.known_absent.update(known_absent)
self.cache[key] = entry
def _insert(self, key, value, known_absent):
self.cache[key] = DictionaryEntry(True, known_absent, value)
class DictionaryCache(object):
"""Caches key -> dictionary lookups, supporting caching partial dicts, i.e.
fetching a subset of dictionary keys for a particular key.
"""
def __init__(self, name, max_entries=1000):
self.cache = LruCache(max_size=max_entries)
self.name = name
self.sequence = 0
self.thread = None
# caches_by_name[name] = self.cache
class Sentinel(object):
__slots__ = []
self.sentinel = Sentinel()
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, dict_keys=None):
entry = self.cache.get(key, self.sentinel)
if entry is not self.sentinel:
cache_counter.inc_hits(self.name)
if dict_keys is None:
return DictionaryEntry(entry.full, dict(entry.value))
else:
return DictionaryEntry(entry.full, {
k: entry.value[k]
for k in dict_keys
if k in entry.value
})
cache_counter.inc_misses(self.name)
return DictionaryEntry(False, {})
def invalidate(self, key):
self.check_thread()
# 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)
def invalidate_all(self):
self.check_thread()
self.sequence += 1
self.cache.clear()
def update(self, sequence, key, value, full=False):
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)
if full:
self._insert(key, value)
else:
self._update_or_insert(key, value)
def _update_or_insert(self, key, value):
entry = self.cache.setdefault(key, DictionaryEntry(False, {}))
entry.value.update(value)
def _insert(self, key, value):
self.cache[key] = DictionaryEntry(True, value)
class DictionaryCache(object):
"""Caches key -> dictionary lookups, supporting caching partial dicts, i.e.
fetching a subset of dictionary keys for a particular key.
"""
def __init__(self, name, max_entries=1000):
self.cache = LruCache(max_size=max_entries, size_callback=len)
self.name = name
self.sequence = 0
self.thread = None
# caches_by_name[name] = self.cache
class Sentinel(object):
__slots__ = []
self.sentinel = Sentinel()
self.metrics = register_cache(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, dict_keys=None):
"""Fetch an entry out of the cache
Args:
key
dict_key(list): If given a set of keys then return only those keys
that exist in the cache.
Returns:
DictionaryEntry
"""
entry = self.cache.get(key, self.sentinel)
if entry is not self.sentinel:
self.metrics.inc_hits()
if dict_keys is None:
return DictionaryEntry(entry.full, entry.known_absent, dict(entry.value))
else:
return DictionaryEntry(entry.full, entry.known_absent, {
k: entry.value[k]
for k in dict_keys
if k in entry.value
})
self.metrics.inc_misses()
return DictionaryEntry(False, set(), {})
def invalidate(self, key):
self.check_thread()
# 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)
def invalidate_all(self):
self.check_thread()
self.sequence += 1
self.cache.clear()
def update(self, sequence, key, value, full=False, known_absent=None):
"""Updates the entry in the cache
Args:
sequence
key
value (dict): The value to update the cache with.
full (bool): Whether the given value is the full dict, or just a
partial subset there of. If not full then any existing entries
for the key will be updated.
known_absent (set): Set of keys that we know don't exist in the full
dict.
"""
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)
if known_absent is None:
known_absent = set()
if full:
self._insert(key, value, known_absent)
else:
self._update_or_insert(key, value, known_absent)
def _update_or_insert(self, key, value, known_absent):
# We pop and reinsert as we need to tell the cache the size may have
# changed
entry = self.cache.pop(key, DictionaryEntry(False, set(), {}))
entry.value.update(value)
entry.known_absent.update(known_absent)
self.cache[key] = entry
def _insert(self, key, value, known_absent):
self.cache[key] = DictionaryEntry(True, known_absent, value)
class Cache(object):
__slots__ = (
"cache",
"max_entries",
"name",
"keylen",
"sequence",
"thread",
"metrics",
)
def __init__(self, name, max_entries=1000, keylen=1, tree=False):
cache_type = TreeCache if tree else dict
self.cache = LruCache(
max_size=max_entries, keylen=keylen, cache_type=cache_type
)
self.name = name
self.keylen = keylen
self.sequence = 0
self.thread = None
self.metrics = register_cache(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, callback=None):
val = self.cache.get(key, _CacheSentinel, callback=callback)
if val is not _CacheSentinel:
self.metrics.inc_hits()
return val
self.metrics.inc_misses()
if default is _CacheSentinel:
raise KeyError()
else:
return default
def update(self, sequence, key, value, callback=None):
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, callback=callback)
def prefill(self, key, value, callback=None):
self.cache.set(key, value, callback=callback)
def invalidate(self, key):
self.check_thread()
if not isinstance(key, tuple):
raise TypeError(
"The cache key must be a tuple not %r" % (type(key),)
)
# 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)
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)
def invalidate_all(self):
self.check_thread()
self.sequence += 1
self.cache.clear()
