xiaoxia-saas/apps/worker/worker_app/tasks/atom_clips.py

"""素材原子切片 Celery 任务 — #1970 智能剪辑流程重构 P1.

素材入库预处理完成（ingest 置 READY）后异步触发：
根据素材时长和已缓存的 scdet 切换点计算原子片段并落库。
失败不阻断素材入库主流程（atom_clips 未就绪时选片有内存兜底）。

P2 增强：切片完成后自动链式触发 AI 标签任务（每个 clip 一个 tag_atom_clip 任务）。
"""

from __future__ import annotations

from celery.utils.log import get_task_logger
from worker_app.celery_app import celery_app
from worker_app.db import SessionLocal

from packages.adapters.sqlalchemy_impl.asset_atom_clip_repository import (
    SQLAlchemyAssetAtomClipRepository,
)
from packages.adapters.sqlalchemy_impl.asset_repository import SQLAlchemyAssetRepository
from packages.domain.atom_clip_service import compute_atom_clips
from packages.domain.plan_generator_utils import extract_scene_points_from_metadata

logger = get_task_logger(__name__)


@celery_app.task(name="worker.generate_atom_clips")
def generate_atom_clips(asset_id: str) -> dict:
    """为单条视频素材生成原子片段。

    Returns:
        任务结果 dict：status / asset_id / clips_count。
    """
    db = SessionLocal()
    try:
        asset_repo = SQLAlchemyAssetRepository(db)
        atom_repo = SQLAlchemyAssetAtomClipRepository(db)

        asset = asset_repo.find_by_id(asset_id)
        if asset is None:
            return {"status": "skipped", "reason": "asset not found", "asset_id": asset_id}

        # 仅视频素材切片
        if asset.mime_type and not asset.mime_type.startswith("video/"):
            return {"status": "skipped", "reason": "not a video", "asset_id": asset_id}
        if not asset.duration or asset.duration <= 0:
            return {"status": "skipped", "reason": "invalid duration", "asset_id": asset_id}

        # 已生成过则幂等跳过（重新切片需先显式删除）
        existing = atom_repo.count_by_asset(asset_id)
        if existing > 0:
            return {
                "status": "skipped",
                "reason": "already generated",
                "asset_id": asset_id,
                "clips_count": existing,
            }

        scene_points = extract_scene_points_from_metadata(asset.metadata)
        # P1 阶段继承素材的标签 ID；片段级语义标签是 P2 功能
        tags = list(getattr(asset, "tag_ids", []) or [])

        clips = compute_atom_clips(
            asset_id=asset_id,
            duration=float(asset.duration),
            scene_change_points=scene_points,
            tags=tags,
        )
        if not clips:
            return {"status": "skipped", "reason": "no clips computed", "asset_id": asset_id}

        atom_repo.batch_create(clips)
        logger.info(
            "[atom_clips] asset_id=%s 生成 %d 个原子片段",
            asset_id,
            len(clips),
        )

        # P2 增强：链式触发 AI 标签任务（每个 clip 一个异步任务）
        _dispatch_tagging_tasks(clips)

        return {"status": "completed", "asset_id": asset_id, "clips_count": len(clips)}
    except Exception as exc:  # noqa: BLE001 - 后台任务兜底，失败不阻断主流程
        db.rollback()
        logger.exception("[atom_clips] asset_id=%s 生成失败: %s", asset_id, exc)
        return {"status": "failed", "asset_id": asset_id, "error": str(exc)}
    finally:
        db.close()


def _dispatch_tagging_tasks(clips: list) -> None:
    """为每个新建片段发送 AI 标签异步任务.

    失败不阻断（标签任务是锦上添花，不影响核心流程）。
    """
    try:
        for clip in clips:
            celery_app.send_task(
                "worker.tag_atom_clip",
                args=[clip.id],
            )
        logger.info(
            "[atom_clips] 已发送 %d 个 AI 标签任务",
            len(clips),
        )
    except Exception as e:
        logger.warning(
            "[atom_clips] 发送 AI 标签任务失败（不影响切片结果）: %s",
            e,
        )