FishServer/fish_api/app/settings.py

from __future__ import annotations

from functools import lru_cache
from pathlib import Path
from typing import Optional

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


def _fish_api_env_file() -> Path:
    """fish_api/.env — 与启动 cwd 无关，避免从仓库根跑 uvicorn 时读不到 .env。"""
    return Path(__file__).resolve().parents[1] / ".env"


def fish_repo_root() -> Path:
    # fish_api/app/settings.py -> parent[2] = repo root (contains FishMeasure/, fish_api/)
    return Path(__file__).resolve().parents[2]


def models_dir() -> Path:
    """仓库内统一权重目录（YOLO / DGCNN / PointNet / X3D / SAM 等），与 FishMeasure 代码目录解耦。"""
    return fish_repo_root() / "models"


def _default_stream_tmp() -> Path:
    return fish_repo_root() / "fish_api" / ".data" / "ingest"


def _default_media_root() -> Path:
    return fish_repo_root() / "fish_api" / ".data" / "media"


def _default_sqlite_path() -> Path:
    return fish_repo_root() / "fish_api" / ".data" / "app.db"


def _default_action_output_root() -> Path:
    return fish_repo_root() / "fish_api" / ".data" / "action_output"


class Settings(BaseSettings):
    model_config = SettingsConfigDict(
        env_file=_fish_api_env_file(),
        env_file_encoding="utf-8",
        extra="ignore",
    )

    #: 对外可访问的 API 基址（无末尾 `/`），用于 biomass 等 JSON 里 `video_left` / `video_right` 的绝对 URL。环境变量：**PUBLIC_BASE_URL**
    public_base_url: str = Field(
        default="http://127.0.0.1:8000",
        validation_alias=AliasChoices("PUBLIC_BASE_URL", "public_base_url"),
    )

    ingest_api_key: str = ""

    stream_tmp_dir: Path = Field(default_factory=_default_stream_tmp)
    media_root: Path = Field(default_factory=_default_media_root)
    sqlite_path: Path = Field(default_factory=_default_sqlite_path)

    fish_measure_root: Path = fish_repo_root() / "FishMeasure"
    fish_action_root: Path = fish_repo_root() / "FishAction"

    #: FishMeasure 推理输出（与 SQLite、媒体缓存同属 fish_api/.data，启动脚本会清空）
    measure_output_root: Path = fish_repo_root() / "fish_api" / ".data" / "measure_output"
    #: FishAction 侧预留目录（与 measure 对称；当前推理多为临时文件，仍随启动清空）
    action_output_root: Path = Field(default_factory=_default_action_output_root)

    python_fish_measure: str = ""
    python_fish_action: str = ""

    yolo_model: Optional[str] = None
    weight_checkpoint: Optional[str] = None

    sam_device: str = "cuda"
    predict_conf: float = 0.5
    predict_imgsz: int = 640
    predict_max_frames: int = 0
    predict_frame_stride: int = 1

    #: 传给 predict_weigth_from_svo2.py 的点云/权重选项（与命令行一致，可用 .env 覆盖）
    predict_filter_pointcloud: bool = True
    predict_use_density_filter: bool = True
    predict_use_clustering_filter: bool = False
    #: 留空则在 _default_paths 中设为 FishMeasure 下默认 PointNet++ 权重（若文件存在）
    predict_pointcloud_classifier: Optional[str] = None
    predict_use_pointcloud_classifier: bool = True
    predict_pointcloud_classifier_threshold: float = 0.7
    predict_use_flatness_filter: bool = True
    predict_flatness_threshold: float = 55.0
    measure_weight_top_k: int = 5
    measure_weight_top_by_length: bool = True
    #: 为 False 时向 predict 传 --no-reuse-existing-clouds，每次强制跑 fish_video（避免误用空/陈旧 cloud；可设 True 加速重复跑同一 SVO）
    measure_reuse_existing_clouds: bool = False
    #: 为 True 时 fish_video 内联 DGCNN + 预览叠加（更重；需 fish_video 已支持）
    predict_fish_video_weight_overlay: bool = False
    predict_minute_interval_sec: float = 60.0

    action_checkpoint: Optional[str] = None
    action_clips_per_video: int = 8
    action_batch_size: int = 4
    action_num_workers: int = 2

    #: 非空时由 fish_api 在后台持续扫描该目录中的新 MP4 并跑 FishAction（与 ingest 共用 SQLite 最新结果）
    action_watch_dir: Optional[Path] = None
    action_watch_poll_interval: float = Field(default=2.0, ge=0.1)
    action_watch_stable_polls: int = Field(default=3, ge=1)
    action_watch_recursive: bool = False
    #: 默认：<action_watch_dir>/.fishaction_watch_processed.json
    action_watch_state_file: Optional[Path] = None
    action_watch_use_state_file: bool = True

    #: 非空时后台持续扫描该目录中的新 .svo2 并跑 FishMeasure（与 ingest 共用 SQLite 最新结果）
    measure_watch_dir: Optional[Path] = None
    measure_watch_poll_interval: float = Field(default=2.0, ge=0.1)
    measure_watch_stable_polls: int = Field(default=3, ge=1)
    measure_watch_recursive: bool = False
    measure_watch_state_file: Optional[Path] = None
    measure_watch_use_state_file: bool = True

    default_fish_species: str = "大黄鱼"

    @field_validator(
        "action_watch_dir",
        "action_watch_state_file",
        "measure_watch_dir",
        "measure_watch_state_file",
        mode="before",
    )
    @classmethod
    def _empty_str_path_none(cls, v: object) -> object:
        if v is None:
            return None
        if isinstance(v, str) and not v.strip():
            return None
        return v

    @model_validator(mode="after")
    def _default_paths(self) -> "Settings":
        md = models_dir()
        if not self.yolo_model:
            object.__setattr__(self, "yolo_model", str(md / "yolo" / "best.pt"))
        if not self.weight_checkpoint:
            object.__setattr__(
                self, "weight_checkpoint", str(md / "weight_estimator" / "best.pt")
            )
        if not self.action_checkpoint:
            object.__setattr__(
                self, "action_checkpoint", str(md / "action_x3d" / "checkpoint_best.pt")
            )
        if not self.predict_pointcloud_classifier:
            _pc = md / "pointcloud" / "best_model.pth"
            if _pc.is_file():
                object.__setattr__(self, "predict_pointcloud_classifier", str(_pc))
        return self


@lru_cache
def get_settings() -> Settings:
    return Settings()