KCjU dZddlZddlZddlmZddlmZddlmZeZ ee Z de de de d e d e f d ZGd d ZeZy)u] Redis-backed result cache. Crop + district + month combinations repeat heavily in real usage — many farmers in the same district asking about the same crop in the same month. Caching the full analysis result avoids re-running 6 models and an LLM call for an answer that hasn't changed. Cache invalidation strategy: time-based TTL (REDIS_CACHE_TTL_SECONDS, default 24h) rather than manual invalidation. This is intentional — climate/market forecasts are inherently "as of today" data, so letting them expire naturally is simpler and safer than trying to track every upstream data change that should invalidate a cache key. If Redis is down, every method here fails open (returns None / does nothing) rather than raising — caching is a performance optimization, not a correctness requirement, and a cache outage must never become a full service outage. N)Redis) get_settings) get_loggerdistrictcropmonth model_versionreturnc|d|d|d|}tj|jdjdd}d|S)u Deterministic cache key. model_version is included so that retraining and deploying new models automatically invalidates old cached results — no manual cache flush needed on model updates. :zutf-8Nz analysis:)hashlibsha256encode hexdigest)rrrr rawdigests GC:\Crop_Prediction\Backend\crop-ai-system\app\services\cache_service.pybuild_cache_keyr!sT JavQugQ}o 6C ^^CJJw/ 0 : : ^>^C++//#tzz%'8S/A A A a NN8=YZ]Y^_ ` ` asABAA!AA!BA!! B*B B BBcK|jy |jjd{y7#t$rYywxYww)NFT)rr'r*rs r health_checkzCacheService.health_checkPsG ;;  ++""$ $ $ %  s0A646A6 AAAAc|jS)zCExposed for the rate limiter, which needs the same connection pool.rrs r raw_clientzCacheService.raw_clientYs{{r)r Nr)__name__ __module__ __qualname__rr-strdictr0intr9boolr=propertyrr?rrrr,sy)  S TD[ aSaaC$JaRVaDEDLrr)__doc__rr1 redis.asynciorapp.core.configrapp.core.logging_configrr%r@r(rCrErr cache_servicerHrrrNsi( (. > H  c  S  QT 00h r