backend/app/core/config.py

"""
Konfiguracja aplikacji.

Wykorzystuje Pydantic Settings do zarządzania zmiennymi środowiskowymi.
"""

from functools import lru_cache

from pydantic_settings import BaseSettings, SettingsConfigDict


class Settings(BaseSettings):
    """
    Ustawienia aplikacji ładowane ze zmiennych środowiskowych.
    """

    model_config = SettingsConfigDict(
        env_file=(".env", "backend/.env"),
        env_file_encoding="utf-8",
        case_sensitive=False,
        extra="ignore"
    )

    # MongoDB
    mongodb_url: str = ""
    mongodb_db_name: str = "sentimentSummary"

    # App Mode
    app_mode: str = "full"  # "full" = monolith, "api" = API-only (no frontend)

    # App Settings
    debug: bool = False
    cors_origins: str = "http://localhost:5173,http://localhost:3000"

    # Cache Settings
    cache_ttl_hours: int = 24
    cache_ttl_short_hours: int = 12      # frequently updated games
    cache_ttl_long_hours: int = 168      # stable games (7 days)
    cache_ttl_worker_managed_hours: int = 1440  # 60 days to preserve stale fallback results
    cache_ttl_on_demand_hours: int = 1440       # 60 days to preserve stale fallback results

    # Incremental Analysis
    incremental_enabled: bool = True
    incremental_max_stored_ids: int = 5000
    incremental_max_gap_days: int = 90  # fall back to full analysis after this many days without reviews
    recent_sample_limit: int = 1000
    niche_cache_max_age_days: int = 60
    analysis_freshness_max_age_days: int = 60
    patch_context_max_age_days: int = 90
    dlc_min_reviews_for_analysis: int = 50
    dlc_visible_in_search: bool = False       # Temporary policy: hide DLC from autocomplete/suggestions
    dlc_worker_analysis_enabled: bool = False  # Temporary policy: exclude DLC from worker-managed analysis

    # Steam API Settings
    review_batch_size: int = 100
    steam_review_language: str = "schinese"  # Review fetch scope; product analyzes Simplified Chinese Steam reviews.
    steam_region: str = "CN"  # CN, US, etc.

    # Steam API Retry
    steam_retry_max_attempts: int = 3
    steam_retry_base_delay: float = 1.0  # doubles each retry
    steam_retry_max_delay: float = 10.0  # cap

    # Steam API Error Cache TTL (seconds)
    steam_error_cache_ttl_404: int = 3600  # 1h
    steam_error_cache_ttl_429: int = 300  # 5min

    # Sampling Settings - Statistical sampling parameters
    sample_top_helpful: int = 50
    sample_confidence_level: float = 0.95
    sample_margin_of_error: float = 0.02
    sample_max_reviews: int = 3000
    sample_minority_min: int = 100

    # NLP Settings - Analysis Parameters
    text_max_length: int = 512
    sentiment_positive_threshold: float = 0.1
    sentiment_negative_threshold: float = -0.1
    topic_min_mentions: int = 5

    # NLP Settings - Deduplication Cache
    dedup_cache_maxsize: int = 10000

    # NLP Settings - Performance & Logic
    nlp_onnx_intra_threads: int = 2
    nlp_onnx_inter_threads: int = 2
    nlp_negation_window: int = 3

    # Prediction Settings
    prediction_retention_threshold_pos: float = 0.2
    prediction_retention_threshold_neg: float = -0.2

    # Community Highlights
    highlights_ngram_min: int = 2
    highlights_ngram_max: int = 5
    highlights_min_mentions: int = 3
    highlights_max_doc_freq_ratio: float = 0.4
    highlights_top_n_general: int = 15
    highlights_top_n_per_topic: int = 5

    # Worker — Pre-cache
    worker_trigger_token: str = ""
    precache_enabled: bool = False
    precache_top_n_games: int = 500
    precache_batch_delay_seconds: int = 10
    precache_checkpoints_hours: str = "6,12,24,72,168,336"
    precache_max_analyses_per_cycle: int = 50

    # Worker — Priority Games
    steam_priority_categories: str = "top_sellers,new_releases,specials"
    steam_priority_regions: str = "CN,US"
    steam_priority_grace_days: int = 3
    steam_priority_categories_url: str = "https://store.steampowered.com/api/featuredcategories"
    steam_bootstrap_max_per_cycle: int = 20
    steam_bootstrap_delay: float = 1.5

    # Worker — News Scan
    news_refresh_window_hours: int = 6
    news_initial_count: int = 20
    news_incremental_count: int = 5

    # Worker — Game Sync
    game_sync_enabled: bool = False
    game_sync_steamspy_delay: float = 61.0
    game_sync_details_delay: float = 1.1
    game_sync_top_n_details: int = 500
    game_sync_cn_enrichment_delay: float = 1.5
    game_sync_cn_enrichment_limit: int = 200
    game_sync_app_type_enrichment_delay: float = 1.5
    game_sync_app_type_enrichment_limit: int = 200

    # Logging (both Live API and Worker)
    worker_log_dir: str = "/data/worker_logs"
    worker_log_fallback_dir: str = "/tmp/worker_logs"
    worker_log_max_bytes: int = 5_000_000       # 5 MB per file
    worker_log_backup_count: int = 3            # 3 rotated files = 20 MB max
    nlp_verbose_logging: bool = False           # re-enable NLP debug logs to stdout
    nlp_debug_log_max_bytes: int = 2_000_000    # 2 MB per file
    errors_log_max_bytes: int = 2_000_000       # 2 MB per file

    # Rate Limiting
    rate_limit_analyze: str = "10/minute"
    rate_limit_default: str = "30/minute"

    # NLP Settings - Hugging Face Models
    # Using specialized Chinese model (RoBERTa-JD) - 90% accuracy on product reviews
    hf_sentiment_model: str = "uer/roberta-base-finetuned-jd-binary-chinese"

    @property
    def cors_origins_list(self) -> list[str]:
        """Zwraca listę dozwolonych originów CORS."""
        return [origin.strip() for origin in self.cors_origins.split(",")]

    @property
    def precache_checkpoints_list(self) -> list[int]:
        """Parse checkpoint hours from comma-separated string."""
        return sorted(int(h.strip()) for h in self.precache_checkpoints_hours.split(","))

    @property
    def steam_priority_categories_list(self) -> list[str]:
        return [c.strip() for c in self.steam_priority_categories.split(",") if c.strip()]

    @property
    def steam_priority_regions_list(self) -> list[str]:
        return [r.strip() for r in self.steam_priority_regions.split(",") if r.strip()]


@lru_cache
def get_settings() -> Settings:
    """Zwraca singleton instancji Settings."""
    return Settings()


settings = get_settings()