diary-news/backend/app/config.py

"""应用配置:从 .env / 环境变量读取,集中管理所有开关。"""
from __future__ import annotations

from functools import lru_cache
from pathlib import Path

from pydantic import Field, field_validator
from pydantic_settings import BaseSettings, SettingsConfigDict


class Settings(BaseSettings):
    model_config = SettingsConfigDict(
        env_file=".env",
        env_file_encoding="utf-8",
        case_sensitive=False,
        extra="ignore",
    )

    # ===== 通用 =====
    tz: str = "Asia/Hong_Kong"
    log_level: str = "INFO"

    # ===== 数据库 =====
    postgres_user: str
    postgres_password: str
    postgres_db: str
    postgres_host: str = "postgres"
    postgres_port: int = 5432

    @property
    def database_url(self) -> str:
        # asyncpg
        return (
            f"postgresql+asyncpg://{self.postgres_user}:{self.postgres_password}"
            f"@{self.postgres_host}:{self.postgres_port}/{self.postgres_db}"
        )

    @property
    def sync_database_url(self) -> str:
        # alembic 用的同步 URL
        return (
            f"postgresql+psycopg2://{self.postgres_user}:{self.postgres_password}"
            f"@{self.postgres_host}:{self.postgres_port}/{self.postgres_db}"
        )

    # ===== Redis =====
    redis_host: str = "redis"
    redis_port: int = 6379
    redis_password: str
    redis_db: int = 0

    @property
    def redis_url(self) -> str:
        return (
            f"redis://:{self.redis_password}@{self.redis_host}:{self.redis_port}/{self.redis_db}"
        )

    # ===== JWT =====
    jwt_secret: str
    jwt_algorithm: str = "HS256"
    access_token_ttl_min: int = 60
    refresh_token_ttl_day: int = 14

    # ===== 腾讯云 TMT =====
    tencentcloud_secret_id: str = ""
    tencentcloud_secret_key: str = ""
    tencentcloud_region: str = "ap-hongkong"
    tencent_tmt_endpoint: str = "tmt.tencentcloudapi.com"
    tencent_tmt_quota_month: int = 5_000_000
    tencent_tmt_quota_buffer: float = 0.05
    tencent_tmt_max_chars_per_req: int = 4500

    # ===== 智谱 GLM(OpenAI 兼容,翻译主通道)=====
    # 用法:智谱开放平台 GLM-4 系列,通过 OpenAI 协议调用
    # 留空 api_key = 不启用该 provider
    zhipu_api_key: str = ""
    zhipu_base_url: str = "https://open.bigmodel.cn/api/paas/v4"
    zhipu_chat_model: str = "glm-4-flash"
    zhipu_model: str = "glm-4-flash"  # 兼容旧字段名
    # 2 秒/次(用户要求 6/11,降低频率避免触发限流)
    zhipu_interval_sec: float = 2.0

    # ===== 讯飞星火(WebSocket,翻译二级通道)=====
    # 用法:讯飞星火 v1.1 Spark Lite,WebSocket 鉴权需要 APPID + APIKey + APISecret
    # 留空任意一个 = 不启用该 provider
    # (历史实现用的是 OpenAI 兼容 + APIPassword,已切换为 WebSocket 鉴权,字段名换)
    spark_appid: str = ""
    spark_api_key: str = ""  # WebSocket 鉴权用的 APIKey
    spark_api_secret: str = ""  # WebSocket 鉴权用的 APISecret
    spark_domain: str = "lite"  # v1.1 Spark Lite
    # 兼容旧字段名(留空,只在没填 WebSocket 字段时起提示作用)
    spark_api_password: str = ""
    # 2 秒/次(用户要求 6/11,降低频率避免触发限流)
    spark_interval_sec: float = 2.0

    @field_validator("tencent_tmt_quota_buffer")
    @classmethod
    def _check_buffer(cls, v: float) -> float:
        if not 0.0 <= v <= 0.5:
            raise ValueError("buffer 必须在 0~0.5")
        return v

    # ===== 本地翻译 =====
    local_translate_enabled: bool = False
    local_translate_model: str = "nllb-200-distilled-600M"
    local_translate_device: str = "cpu"

    # ===== 腾讯 MaaS(OpenAI 兼容翻译备用通道)=====
    # 用法:腾讯云 MaaS 提供的翻译模型,通过 OpenAI 协议调用
    # 留空 api_key = 不启用该 provider
    tencent_maas_api_key: str = ""
    tencent_maas_base_url: str = "https://maas-api.hivoice.cn/v1"
    tencent_maas_model: str = "u2"
    # 2 秒/次(与智谱/星火统一节流)
    tencent_maas_interval_sec: float = 2.0

    # ===== 抓取 =====
    fetch_global_qps: int = 4
    fetch_timeout: int = 20
    fetch_fail_pause_threshold: int = 3
    fetch_max_retries: int = 2

    # ===== API Push 短新闻 ingest 限速 =====
    # 每个 ingest token 的滑动窗口限速(篇/秒)。2 = 短新闻一秒最多推 2 篇
    # 单 token — 防止单点滥用。改小需重启 api。
    ingest_rate_per_sec: int = 2

    # ===== 站点并发登录 IP 限制 =====
    # 限制同时在线的客户端 IP 数(防滥用 + 防 token 泄漏被滥用)
    # Redis ZSET 滑动窗口:每次已认证请求刷新 score,30 天没活动自动剔除
    # 第 (limit+1) 个新 IP 登录时直接 429
    site_max_active_ips: int = 30
    site_active_ip_idle_days: int = 30
    # 是否信任反向代理的 X-Forwarded-For 头(生产用 Caddy/Nginx 必开;直连调测关)
    trust_x_forwarded_for: bool = True

    # ===== Caddy / 域名 =====
    domain: str = ""
    acme_email: str = ""

    # ===== Agnes LLM(智能增强)=====
    # 留空 = 不启用 LLM 增强(翻译后只走默认排版,提示词也不读)
    agnes_api_key: str = ""
    agnes_base_url: str = "https://apihub.agnes-ai.com/v1"
    agnes_chat_model: str = "agnes-2.0-flash"
    agnes_image_model: str = "agnes-image-2.1-flash"
    # 全局 LLM 调用间隔(秒),避免被限流
    llm_interval_sec: float = 2.0

    # ===== 美团大模型 LongCat(双 provider 评论的第二个)=====
    # OpenAI 兼容端点;与 Agnes 并列,各自跑各自的 prompt,结果存到 articles 各自列
    # 留空 api_key = 不启用美团 provider(Angel 仍正常工作)
    meituan_api_key: str = ""
    meituan_base_url: str = "https://api.longcat.chat/openai/v1"
    meituan_chat_model: str = "LongCat-2.0-Preview"
    meituan_interval_sec: float = 2.0

    # ===== 内部路径(部署后可调) =====
    project_root: Path = Path(__file__).resolve().parents[2]


@lru_cache
def get_settings() -> Settings:
    return Settings()  # type: ignore[call-arg]


settings = get_settings()