rmi-backend/app/domains/databus/cache.py

409 lines
17 KiB
Python

"""
DataBus Cache Layer - Three-Tier Cache with SWR + Per-Type Stats
================================================================
L1: In-memory dict (sub-millisecond, 4096 keys, LRU eviction) + SWR stale buffer
L2: Redis (sub-millisecond, shared across processes, TTL-managed)
L3: Cloudflare R2 cold storage (RAG permanence, nightly snapshots)
Stale-While-Revalidate (SWR):
- L1 stores both fresh and stale entries (stale = TTL * 2)
- On cache read, if entry is fresh → direct hit
- If entry is stale (past TTL but within stale window) → return stale data
AND flag for background refresh via cache.stale_refresh_callback
- User NEVER waits for a refresh - always gets instant data
Per-Type Stats:
- Tracks hits/misses per data_type for tuning TTLs
- health() flags types with hit_rate < 30% as "increase TTL"
"""
import asyncio
import contextlib
import hashlib
import json
import logging
import os
import time
from collections import OrderedDict, defaultdict
from collections.abc import Callable
from typing import Any
from dotenv import load_dotenv
load_dotenv("/app/.env", override=True)
logger = logging.getLogger("databus.cache")
REDIS_HOST = os.getenv("REDIS_HOST", "rmi-redis")
REDIS_PORT = int(os.getenv("REDIS_PORT", "6379"))
REDIS_PASSWORD = os.getenv("REDIS_PASSWORD", "")
REDIS_DB = int(os.getenv("REDIS_DB", "0"))
class L1Cache:
"""In-memory LRU cache with Stale-While-Revalidate support.
Entries are stored with TWO expiry windows:
- fresh_expiry: data is fresh, return immediately (normal hit)
- stale_expiry: data is stale but usable (SWR hit)
Stale entries are returned instantly; the caller triggers background refresh.
"""
def __init__(self, max_keys: int = 4096, stale_multiplier: float = 2.5):
self._max = max_keys
self._stale_mult = stale_multiplier
self._store: OrderedDict[str, tuple[Any, float, float]] = OrderedDict()
# (value, fresh_expiry, stale_expiry)
self.hits = 0
self.stale_hits = 0
self.misses = 0
self._lock = asyncio.Lock()
async def get(self, key: str) -> tuple[Any | None, bool]:
"""Returns (value, is_stale). is_stale=True means background refresh needed."""
async with self._lock:
entry = self._store.get(key)
if entry is None:
self.misses += 1
return None, False
value, fresh_expiry, stale_expiry = entry
now = time.monotonic()
if now > stale_expiry:
# Fully expired - evict
del self._store[key]
self.misses += 1
return None, False
if now > fresh_expiry:
# Stale but usable - SWR hit
self._store.move_to_end(key)
self.stale_hits += 1
return value, True
# Fresh hit
self._store.move_to_end(key)
self.hits += 1
return value, False
async def set(self, key: str, value: Any, ttl: int):
async with self._lock:
now = time.monotonic()
fresh = now + ttl
stale = now + int(ttl * self._stale_mult)
self._store[key] = (value, fresh, stale)
self._store.move_to_end(key)
while len(self._store) > self._max:
self._store.popitem(last=False)
async def delete(self, key: str):
async with self._lock:
self._store.pop(key, None)
async def clear(self):
async with self._lock:
self._store.clear()
def stats(self) -> dict:
total = self.hits + self.stale_hits + self.misses
return {
"keys": len(self._store),
"max_keys": self._max,
"hits": self.hits,
"stale_hits": self.stale_hits,
"misses": self.misses,
"hit_rate": round((self.hits + self.stale_hits) / total * 100, 1) if total > 0 else 0,
"fresh_hit_rate": round(self.hits / total * 100, 1) if total > 0 else 0,
}
class L2RedisCache:
"""Redis cache. Shared across processes. TTL-managed automatically."""
def __init__(self):
self._redis = None
self._available = False
self._prefix = "databus:"
self.hits = 0
self.misses = 0
async def _connect(self):
if self._redis and self._available:
return True
try:
import redis.asyncio as aioredis
kwargs = {
"host": REDIS_HOST,
"port": REDIS_PORT,
"db": REDIS_DB,
"socket_connect_timeout": 2,
"socket_timeout": 2,
"decode_responses": True,
"protocol": 2, # Redis 7.2 compat - avoid HELLO/AUTH handshake issue
}
if REDIS_PASSWORD:
kwargs["password"] = REDIS_PASSWORD
self._redis = aioredis.Redis(**kwargs)
await self._redis.ping()
self._available = True
logger.info("DataBus Cache: Redis connected")
return True
except Exception as e:
logger.warning(f"DataBus Cache: Redis unavailable ({e}), L2 disabled")
self._available = False
return False
async def get(self, key: str) -> Any | None:
if not self._available and not await self._connect():
self.misses += 1
return None
try:
raw = await self._redis.get(f"{self._prefix}{key}")
if raw:
self.hits += 1
return json.loads(raw)
self.misses += 1
return None
except Exception:
self._available = False
self.misses += 1
return None
async def set(self, key: str, value: Any, ttl: int):
if not self._available and not await self._connect():
return
try:
await self._redis.setex(f"{self._prefix}{key}", ttl, json.dumps(value, default=str))
except Exception:
self._available = False
async def delete(self, key: str):
if not self._available:
return
with contextlib.suppress(Exception):
await self._redis.delete(f"{self._prefix}{key}")
async def clear(self):
if not self._available:
return
try:
async for key in self._redis.scan_iter(f"{self._prefix}*"):
await self._redis.delete(key)
except Exception:
pass
def stats(self) -> dict:
total = self.hits + self.misses
return {
"available": self._available,
"hits": self.hits,
"misses": self.misses,
"hit_rate": round(self.hits / total * 100, 1) if total > 0 else 0,
}
class CacheLayer:
"""
Three-tier cache with Stale-While-Revalidate + per-type stats.
L1 (memory, SWR) → L2 (Redis) → L3 (R2, async, background)
Read path: L1 (fresh? → done. stale? → return stale + schedule refresh) → L2 → miss
Write path: External API → L1 + L2 (L3 batched via RAG permanence cron)
SWR callback: When L1 returns stale data, cache fires stale_refresh_callback
so the caller can schedule background re-fetch without blocking the user.
"""
def __init__(self):
self.l1 = L1Cache(max_keys=4096)
self.l2 = L2RedisCache()
self._l3_enabled = True
# SWR: caller sets this to a coroutine factory for background refresh
self.stale_refresh_callback: Callable | None = None
# Per-type hit/miss tracking for TTL tuning
self._type_stats: dict[str, dict[str, int]] = defaultdict(lambda: {"hits": 0, "stale_hits": 0, "misses": 0})
# Data type TTLs - optimized for FREE tier usage to avoid rate limits
# and maximize our 1-month Arkham trial + Alchemy free quota.
self.ttl_config = {
# ── High Frequency (Cache aggressively to save free API calls) ──
"token_price": 60, # Increased from 15s to 60s (better cache reuse for price data)
"market_overview": 60, # Increased from 30s to 60s
"trending": 120, # Increased from 60s to 120s
"market_movers": 60, # Increased from 30s to 60s
"alerts": 30, # Increased from 15s to 30s
# ── Medium Frequency ──
"token_detail": 60, # Increased from 30s
"token_meta": 300, # Increased from 120s (Alchemy free tier)
"wallet_balance": 30, # Increased from 10s (Alchemy free tier)
"wallet_tokens": 300, # Increased from 60s to 300s (reduce misses on uncached provider)
"wallet_pnl": 120, # Increased from 60s
"tx_history": 60, # Increased from 30s
"dex_data": 60, # Increased from 30s
"holder_data": 120, # Increased from 60s
# ── Low Frequency (Static or slow-moving data) ──
"risk_scan": 600, # Increased from 300s
"sentinel_deep": 600, # Increased from 300s
"funding_source": 7200, # Increased from 3600s (2 hours)
"solana_funding": 7200, # Increased from 3600s (2 hours)
"wallet_labels": 86400, # 24 hours (local data, no API cost)
"entity_intel": 3600, # Increased from 1800s (1 hour)
"socialfi_resolve": 86400, # 24 hours
"cross_chain": 3600, # Increased from 1800s
"wallet_cluster": 3600, # Increased from 1800s
"bundle_detect": 600, # Increased from 300s
# ── ARKHAM INTELLIGENCE (1-Month Free Trial: 100k quota) ──
# Cache aggressively to maximize the 100,000 monthly request limit
"arkham_entity": 600, # Increased from 300s (10 mins)
"arkham_portfolio": 300, # Increased from 120s (5 mins)
"arkham_labels": 7200, # Increased from 3600s (2 hours)
"arkham_transfers": 300, # Increased from 60s (5 mins)
"arkham_counterparties": 600, # Increased from 300s (10 mins)
# ── Other ──
"nansen_labels": 3600,
"nansen_smart_money": 1800,
"news": 600, # Increased from 300s (10 mins)
"news_intel": 600, # 10 mins for aggregated news
"messari_news": 900, # 15 mins for Messari institutional feed
"social_feed": 300, # Increased from 120s (5 mins)
"sentiment": 600, # Increased from 300s
"whale_data": 300, # Increased from 120s
"smart_money": 300, # Increased from 120s
"gmgn_smart_money": 300, # Increased from 120s
"launches": 120, # Increased from 60s
"bubble_map": 600, # Increased from 300s
"rugmaps_analysis": 1200, # Increased from 600s (20 mins)
"contract_scan": 3600, # Increased from 1800s (1 hour)
"threat_check": 600, # Increased from 300s
"prediction_markets": 120, # Increased from 60s
"prediction_signals": 300, # Increased from 120s
"defi_protocols": 600, # Increased from 300s
"rag_search": 600, # Increased from 300s
"tvl": 300, # Increased from 120s
"wallet_profile": 600, # Increased from 300s
"portfolio": 120, # Increased from 60s
# ── NEW FREE SOURCES ADDED ──
"defillama_tvl": 3600, # 1 hour (completely free, no API key)
"defillama_chains": 3600, # 1 hour (completely free, no API key)
"blockchair_address": 600, # 10 mins (free tier, no API key)
"blockchair_stats": 1800, # 30 mins (free tier, no API key)
"birdeye_overview": 120, # 2 mins (freemium, 50k/mo quota)
"birdeye_price": 30, # 30s (freemium, 50k/mo quota, fallback to DexScreener)
"solana_tracker_price": 15, # 15s (freemium, 5k/mo quota)
"solana_tracker_token": 60, # 1 min (freemium, 5k/mo quota)
"solana_tracker_trending": 120, # 2 mins (freemium, 5k/mo quota)
"dev_activity": 3600, # 1 hour (Santiment free tier, 100 calls/day)
"url_security_scan": 86400, # 24 hours (VirusTotal free tier, 500 req/day)
"dune_early_buyers": 14400, # 4 hours (Dune free tier: 10k CU/mo, aggressive caching)
"default": 60,
}
def _extract_data_type(self, key: str) -> str:
"""Extract data_type from cache key format 'source:data_type:hash'."""
parts = key.split(":")
if len(parts) >= 2:
return parts[1]
return "default"
async def get(self, key: str, data_type: str = "default") -> tuple[Any | None, bool]:
"""Get from cache with SWR. Returns (value, is_stale).
If is_stale=True, caller should schedule background refresh.
"""
# L1 (with SWR)
val, is_stale = await self.l1.get(key)
if val is not None:
dtype = data_type or self._extract_data_type(key)
if is_stale:
self._type_stats[dtype]["stale_hits"] += 1
else:
self._type_stats[dtype]["hits"] += 1
# SWR: schedule background refresh if stale and callback is set
if is_stale and self.stale_refresh_callback:
try: # noqa: SIM105
asyncio.create_task(self.stale_refresh_callback(key))
except Exception:
pass # Best-effort refresh
return val, is_stale
# L2 (no SWR - Redis handles its own TTL)
val = await self.l2.get(key)
if val is not None:
# Promote to L1 with shorter TTL (fresh only for now)
await self.l1.set(key, val, 60)
dtype = data_type or self._extract_data_type(key)
self._type_stats[dtype]["hits"] += 1
return val, False
dtype = data_type or self._extract_data_type(key)
self._type_stats[dtype]["misses"] += 1
return None, False
async def set(self, key: str, value: Any, ttl: int | None = None, data_type: str = "default"):
"""Set in cache. Writes to L1 + L2."""
if ttl is None:
ttl = self.ttl_config.get(data_type, 60)
await self.l1.set(key, value, ttl)
await self.l2.set(key, value, ttl)
async def delete(self, key: str):
await self.l1.delete(key)
await self.l2.delete(key)
async def clear(self):
await self.l1.clear()
await self.l2.clear()
def make_key(self, source: str, data_type: str, **kwargs) -> str:
"""Generate a deterministic cache key."""
args_str = json.dumps(kwargs, sort_keys=True, default=str)
args_hash = hashlib.sha256(args_str.encode()).hexdigest()[:16]
return f"{source}:{data_type}:{args_hash}"
def type_stats(self) -> dict[str, dict]:
"""Per-data-type hit/miss/stale stats with TTL tuning suggestions."""
result = {}
for dtype, counts in self._type_stats.items():
total = counts["hits"] + counts["stale_hits"] + counts["misses"]
hit_rate = round((counts["hits"] + counts["stale_hits"]) / total * 100, 1) if total > 0 else 0
entry = {
"hits": counts["hits"],
"stale_hits": counts["stale_hits"],
"misses": counts["misses"],
"hit_rate": hit_rate,
"current_ttl": self.ttl_config.get(dtype, 60),
}
# Suggest TTL increase if hit_rate < 30%
if total > 10 and hit_rate < 30:
entry["suggestion"] = "increase_ttl"
# Suggest TTL decrease if hit_rate > 95% and TTL > 120
elif total > 10 and hit_rate > 95 and self.ttl_config.get(dtype, 60) > 120:
entry["suggestion"] = "decrease_ttl"
result[dtype] = entry
return result
async def health(self) -> dict:
l1 = self.l1.stats()
l2 = self.l2.stats()
total_hits = l1["hits"] + l1["stale_hits"] + l2["hits"]
total_misses = l1["misses"] + l2["misses"]
total = total_hits + total_misses
return {
"status": "ok",
"l1_memory": l1,
"l2_redis": l2,
"l3_r2": {"enabled": self._l3_enabled, "note": "Batched via RAG permanence cron"},
"combined_hit_rate": round(total_hits / total * 100, 1) if total > 0 else 0,
"total_hits": total_hits,
"total_misses": total_misses,
"per_type_stats": self.type_stats(),
"ttl_config": self.ttl_config,
}
# ── Singleton ─────────────────────────────────────────────────────────────────
_cache: CacheLayer | None = None
def get_cache() -> CacheLayer:
global _cache
if _cache is None:
_cache = CacheLayer()
return _cache