Upate cache.

Browse Source
This commit is contained in:
retoor 2025-09-30 19:35:56 +02:00
parent f5a2928f1b
commit 7b245869d5
1 changed files with 91 additions and 137 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

210
src/snek/system/cache.py
View File

@ -1,167 +1,137 @@
import asyncio
import functools import functools
import json import json
import asyncio from collections import OrderedDict
from collections import OrderedDict # Use OrderedDict for O(1) LRU management
# Assuming snek.system.security exists and security.hash is an async function
from snek.system import security from snek.system import security
# NOTE: functools.cache is only for synchronous functions and is not used in the class
# cache = functools.cache # Unused, removed from class logic
CACHE_MAX_ITEMS_DEFAULT = 5000 CACHE_MAX_ITEMS_DEFAULT = 5000
class Cache: class Cache:
"""
An asynchronous, thread-safe, in-memory LRU (Least Recently Used) cache.
This implementation uses an OrderedDict for efficient O(1) time complexity
for its core get, set, and delete operations.
"""
def __init__(self, app, max_items=CACHE_MAX_ITEMS_DEFAULT): def __init__(self, app, max_items=CACHE_MAX_ITEMS_DEFAULT):
self.app = app self.app = app
# Replaced dict with OrderedDict for O(1) LRU moves # OrderedDict is the core of the LRU logic. It remembers the order
self.cache = OrderedDict() # in which items were inserted.
self.cache: OrderedDict = OrderedDict()
self.max_items = max_items self.max_items = max_items
self.stats = {} self.stats = {}
self.enabled = False self.enabled = True
# LRU list is no longer needed; OrderedDict handles the order # A lock is crucial to prevent race conditions in an async environment.
self.lru = []
# Add an asyncio Lock for concurrent access safety
self._lock = asyncio.Lock() self._lock = asyncio.Lock()
self.version = ((42 + 420 + 1984 + 1990 + 10 + 6 + 71 + 3004 + 7245) ^ 1337) + 4 [cite_start]self.version = ((42 + 420 + 1984 + 1990 + 10 + 6 + 71 + 3004 + 7245) ^ 1337) + 4 [cite: 1760]
# --- Core Cache Logic (Now O(1) operations) --- async def get(self, key):
"""
async def get(self, args): Retrieves an item from the cache. If found, it's marked as recently used.
# Must be protected by a lock for thread safety Returns None if the item is not found or the cache is disabled.
async with self._lock: """
if not self.enabled: if not self.enabled:
return None return None
# Check for cache miss async with self._lock:
if args not in self.cache: if key not in self.cache:
await self.update_stat(args, "get") await self.update_stat(key, "get")
# print("Cache miss!", args, flush=True)
return None return None
await self.update_stat(args, "get") # Mark as recently used by moving it to the end of the OrderedDict.
# This is an O(1) operation.
self.cache.move_to_end(key)
await self.update_stat(key, "get")
return self.cache[key]
# 1. Update LRU order: Move to end (most recently used) async def set(self, key, value):
# Use self.cache.move_to_end() for O(1) LRU update """
value = self.cache.pop(args) # Pop to get the value Adds or updates an item in the cache and marks it as recently used.
self.cache[args] = value # Re-add to the end (MRU) If the cache exceeds its maximum size, the least recently used item is evicted.
"""
# NOTE: The original code had a confusing LRU list implementation
# that was completely wrong. It should have been:
# 1. Check if in self.cache (dict).
# 2. If in cache, move it to the front/end of the LRU structure.
# 3. Return the value.
# Since self.lru is part of the public interface (used in get_stats),
# we must maintain its state for that method, but it is not
# used for core LRU logic anymore.
# print("Cache hit!", args, flush=True)
return value
async def set(self, args, result):
# Must be protected by a lock for thread safety
async with self._lock:
if not self.enabled: if not self.enabled:
return return
is_new = args not in self.cache async with self._lock:
is_new = key not in self.cache
# 1. Update/Set value # Add or update the item. If it exists, it's moved to the end.
self.cache[args] = result self.cache[key] = value
self.cache.move_to_end(key)
# 2. Update LRU order (Move to end/MRU) await self.update_stat(key, "set")
self.cache.move_to_end(args)
await self.update_stat(args, "set") # Evict the least recently used item if the cache is full.
# This is an O(1) operation.
# 3. Handle eviction (Now O(1))
if len(self.cache) > self.max_items: if len(self.cache) > self.max_items:
# popitem(last=False) removes the first (LRU) item # popitem(last=False) removes and returns the first (oldest) item.
evicted_key, _ = self.cache.popitem(last=False) evicted_key, _ = self.cache.popitem(last=False)
# NOTE: The original code failed to update self.lru on eviction. # Optionally, you could log the evicted key here.
# Since we are using OrderedDict, we don't need self.lru for LRU tracking.
# However, if self.lru must be updated for `get_stats`,
# we must manage it here and in `get_stats`.
# For a clean repair, self.cache (OrderedDict) is the source of truth.
if is_new: if is_new:
self.version += 1 self.version += 1
# print(f"Cache store! {len(self.cache)} items. New version:", self.version, flush=True)
async def delete(self, args): async def delete(self, key):
# Must be protected by a lock for thread safety """Removes an item from the cache if it exists."""
async with self._lock:
if not self.enabled: if not self.enabled:
return return
if args in self.cache: async with self._lock:
await self.update_stat(args, "delete") if key in self.cache:
del self.cache[args] await self.update_stat(key, "delete")
# NOTE: No list manipulation needed due to OrderedDict # Deleting from OrderedDict is an O(1) operation on average.
del self.cache[key]
# --- Utility Methods (Interface Retained) ---
async def get_stats(self): async def get_stats(self):
# Must be protected by a lock for thread safety """Returns statistics for all items currently in the cache."""
async with self._lock: async with self._lock:
all_ = [] stats_list = []
# Iterate through self.cache (OrderedDict) to get the MRU-to-LRU order # Items are iterated from oldest to newest. We reverse to show
# The public interface uses self.lru, so we must generate it here # most recently used items first.
# from the source of truth (self.cache keys) in MRU order. for key in reversed(self.cache):
stat_data = self.stats.get(key, {"set": 0, "get": 0, "delete": 0})
value = self.cache[key]
value_record = value.record if hasattr(value, 'record') else value
# Generate the keys in MRU order (reverse of iteration) stats_list.append({
lru_keys = list(self.cache.keys()) "key": key,
# For the original self.lru list, front was MRU, back was LRU "set": stat_data.get("set", 0),
lru_keys.reverse() "get": stat_data.get("get", 0),
self.lru = lru_keys # Update the redundant public attribute self.lru "delete": stat_data.get("delete", 0),
"value": str(self.serialize(value_record)),
})
return stats_list
for key in self.lru: async def update_stat(self, key, action):
"""Updates hit/miss/set counts for a given cache key."""
# This method is already called within a locked context,
# but the lock makes it safe if ever called directly.
async with self._lock:
if key not in self.stats: if key not in self.stats:
self.stats[key] = {"set": 0, "get": 0, "delete": 0} self.stats[key] = {"set": 0, "get": 0, "delete": 0}
self.stats[key][action] += 1
# Handling potential KeyError if key was evicted but stat remains
if key in self.cache:
value_record = self.cache[key].record if hasattr(self.cache.get(key), 'record') else self.cache[key]
all_.append(
{
"key": key,
"set": self.stats[key]["set"],
"get": self.stats[key]["get"],
"delete": self.stats[key]["delete"],
"value": str(self.serialize(value_record)),
}
)
return all_
# Made synchronous as it's a CPU-bound operation
def serialize(self, obj): def serialize(self, obj):
"""A synchronous helper to create a serializable representation of an object."""
if not isinstance(obj, dict):
return obj
cpy = obj.copy() cpy = obj.copy()
cpy.pop("created_at", None) for key_to_remove in ["created_at", "deleted_at", "email", "password"]:
cpy.pop("deleted_at", None) cpy.pop(key_to_remove, None)
cpy.pop("email", None)
cpy.pop("password", None)
return cpy return cpy
# Made synchronous as it's a CPU-bound operation
async def update_stat(self, key, action):
# Although called within locked methods, we lock it here to make it safe
# if called directly, as the original signature is async.
async with self._lock:
if key not in self.stats:
self.stats[key] = {"set": 0, "get": 0, "delete": 0}
self.stats[key][action] = self.stats[key][action] + 1
# Made synchronous as it's a CPU-bound operation
def json_default(self, value): def json_default(self, value):
"""JSON serializer fallback for objects that are not directly serializable."""
try: try:
return json.dumps(value.__dict__, default=str) return json.dumps(value.__dict__, default=str)
except: except:
return str(value) return str(value)
# Retained async due to the call to await security.hash()
async def create_cache_key(self, args, kwargs): async def create_cache_key(self, args, kwargs):
# CPU-bound operations don't need a lock, but retain async for security.hash """Creates a consistent, hashable cache key from function arguments."""
# security.hash is async, so this method remains async.
return await security.hash( return await security.hash(
json.dumps( json.dumps(
{"args": args, "kwargs": kwargs}, {"args": args, "kwargs": kwargs},
@ -171,7 +141,7 @@ class Cache:
) )
def async_cache(self, func): def async_cache(self, func):
# No change to the decorator structure """Decorator to cache the results of an async function."""
@functools.wraps(func) @functools.wraps(func)
async def wrapper(*args, **kwargs): async def wrapper(*args, **kwargs):
cache_key = await self.create_cache_key(args, kwargs) cache_key = await self.create_cache_key(args, kwargs)
@ -184,27 +154,11 @@ class Cache:
return wrapper return wrapper
def async_delete_cache(self, func): def async_delete_cache(self, func):
# The internal logic is now clean O(1) using self.delete() """Decorator to invalidate a cache entry before running an async function."""
@functools.wraps(func) @functools.wraps(func)
async def wrapper(*args, **kwargs): async def wrapper(*args, **kwargs):
cache_key = await self.create_cache_key(args, kwargs) cache_key = await self.create_cache_key(args, kwargs)
# Use the fixed self.delete method
await self.delete(cache_key) await self.delete(cache_key)
return await func(*args, **kwargs) return await func(*args, **kwargs)
return wrapper return wrapper
# --- Standalone async_cache (No Change) ---
# NOTE: This implementation is separate from the Cache class and is not LRU.
def async_cache(func):
cache = {}
@functools.wraps(func)
async def wrapper(*args):
if args in cache:
return cache[args]
result = await func(*args)
cache[args] = result
return result
return wrapper