refactor(transfer): make queue admission durable

app/db/adapters/transfer.py

@@ -0,0 +1,115 @@
+"""整理任务持久准入端口的 SQLAlchemy 适配器。"""
+
+from collections.abc import Callable
+from datetime import datetime
+from uuid import uuid4
+
+from sqlalchemy.exc import IntegrityError
+from sqlalchemy.orm import Session
+
+from app.application.transfer import (
+    TRANSFER_ADMISSION_ACCEPTED,
+    TransferAdmission,
+)
+from app.db.models.transferpending import TransferPending
+from app.db.oper.transferpending import TransferPendingOper
+from app.db.uow import SqlAlchemyUnitOfWork
+
+
+class TransactionalTransferAdmissionRepository:
+    """以短生命周期 Session 实现整理任务持久准入端口。"""
+
+    def __init__(self, session_factory: Callable[[], Session]) -> None:
+        """保存由组合根提供的同步会话工厂。"""
+        self._session_factory = session_factory
+
+    @staticmethod
+    def _now() -> str:
+        """生成与历史登记时间可按字典序比较的当前时间。"""
+        return datetime.now().strftime("%Y-%m-%d %H:%M:%S")
+
+    @staticmethod
+    def _project(pending: TransferPending) -> TransferAdmission:
+        """在 Session 有效期内把 ORM 行冻结为应用层 DTO。"""
+        created_at = pending.created_at or pending.updated_at
+        return TransferAdmission(
+            task_id=pending.task_id,
+            storage=pending.storage,
+            src_path=pending.src_path,
+            state=pending.state,
+            created_at=created_at,
+            updated_at=pending.updated_at,
+            last_error=pending.last_error,
+        )
+
+    def admit(self, *, storage: str, src_path: str) -> TransferAdmission:
+        """幂等持久化准入事实，并返回跨重启稳定的任务标识。"""
+        now_time = self._now()
+        try:
+            with self._session_factory() as session:
+                transaction = SqlAlchemyUnitOfWork(session)
+                try:
+                    pending = TransferPendingOper(db=session).stage_admit(
+                        task_id=uuid4().hex,
+                        storage=storage,
+                        src_path=src_path,
+                        state=TRANSFER_ADMISSION_ACCEPTED,
+                        now_time=now_time,
+                    )
+                    if pending is None:
+                        raise ValueError("整理任务的存储与源路径不能为空")
+                    session.flush()
+                    admission = self._project(pending)
+                    transaction.commit()
+                    return admission
+                except Exception:
+                    transaction.rollback()
+                    raise
+        except IntegrityError as error:
+            # 并发准入可能同时通过查询；唯一约束决定赢家，输家回读稳定身份。
+            with self._session_factory() as session:
+                pending = TransferPendingOper(db=session).get_by_identity(
+                    storage=storage,
+                    src_path=src_path,
+                )
+                if pending is None:
+                    raise RuntimeError("并发准入冲突后未找到已提交记录") from error
+                return self._project(pending)
+
+    def list_accepted(self, limit: int = 5000) -> list[TransferAdmission]:
+        """在独立只读会话中投影等待恢复或执行的准入记录。"""
+        with self._session_factory() as session:
+            pending_items = TransferPendingOper(db=session).list_by_state(
+                state=TRANSFER_ADMISSION_ACCEPTED,
+                limit=limit,
+            )
+            return [self._project(pending) for pending in pending_items]
+
+    def record_enqueue_failure(self, *, task_id: str, error: str) -> None:
+        """独立提交最近一次入队失败，保留准入记录供后续恢复。"""
+        with self._session_factory() as session:
+            transaction = SqlAlchemyUnitOfWork(session)
+            try:
+                TransferPendingOper(db=session).stage_record_enqueue_failure(
+                    task_id=task_id,
+                    error=error,
+                    now_time=self._now(),
+                )
+                transaction.commit()
+            except Exception:
+                transaction.rollback()
+                raise
+
+    def discard_task(self, *, task_id: str) -> int:
+        """在独立事务中按稳定任务标识删除已到终态的准入记录。"""
+        with self._session_factory() as session:
+            transaction = SqlAlchemyUnitOfWork(session)
+            try:
+                deleted = TransferPendingOper(db=session).stage_discard_task(
+                    task_id=task_id,
+                )
+                transaction.commit()
+                return deleted
+            except Exception:
+                transaction.rollback()
+                raise

app/db/models/transferpending.py

@@ -1,6 +1,8 @@
-from typing import List, Optional
+from datetime import datetime
+from typing import List, Optional, cast
+from uuid import uuid4
 
-from sqlalchemy import Index, String, delete, select
+from sqlalchemy import Index, String, Text, UniqueConstraint, delete, select, update
 from sqlalchemy.orm import Mapped, Session, mapped_column
 
 from app.db.base import Base, execute_dml, get_id_column
@@ -15,22 +17,43 @@ class TransferPending(Base):
     蒸发。而已经稳定落地的文件不会再产生任何监控事件，也不会有新的补偿扫描起点
     ——结果就是永久漏件，只能靠人工比对补整理。
 
-    这里只落盘最小事实：存储与源文件路径。重启后重新走一遍整理入口，由整理历史
-    查重挡掉已经完成的，因此不需要序列化 meta/mediainfo 这些重对象，也不存在
-    识别结果陈旧的问题。
+    这里只落盘恢复所需的最小事实：稳定任务身份、存储、源文件路径、准入状态和
+    最近入队错误。重启后重新走一遍整理入口，由整理历史查重挡掉已经完成的，
+    因此不需要序列化 meta/mediainfo 这些重对象，也不存在识别结果陈旧的问题。
     """
 
     id = get_id_column()
+    # 稳定任务标识
+    task_id: Mapped[str] = mapped_column(
+        String(64), nullable=False, default=lambda: uuid4().hex
+    )
     # 存储
     storage: Mapped[str] = mapped_column(String, nullable=False)
     # 源文件路径
     src_path: Mapped[str] = mapped_column(String, nullable=False)
     # 登记时间
     created_at: Mapped[Optional[str]] = mapped_column(String)
+    # 持久状态
+    state: Mapped[str] = mapped_column(String(32), nullable=False, default="accepted")
+    # 最后更新时间
+    updated_at: Mapped[str] = mapped_column(
+        String(40), nullable=False,
+        default=lambda: datetime.now().strftime("%Y-%m-%d %H:%M:%S"),
+    )
+    # 最近一次入队失败原因
+    last_error: Mapped[Optional[str]] = mapped_column(Text)
 
     __table_args__ = (
         # 同一个文件重复入队只保留一条，回放时不会重复送入整理链
         Index("ux_transferpending_storage_path", "storage", "src_path", unique=True),
+        # 恢复主查询按状态过滤、登记时间与主键稳定排序
+        Index(
+            "ix_transferpending_state_created",
+            "state",
+            "created_at",
+            "id",
+        ),
+        UniqueConstraint("task_id", name="uq_transferpending_task_id"),
     )
 
     @classmethod
@@ -50,11 +73,128 @@ class TransferPending(Base):
             select(cls).where(cls.storage == storage, cls.src_path == src_path)
         ).scalars().first()
         if pending:
-            return pending
-        pending = cls(storage=storage, src_path=src_path, created_at=now_time)
+            return cast("TransferPending", pending)
+        pending = cls(
+            storage=storage,
+            src_path=src_path,
+            state="accepted",
+            created_at=now_time,
+            updated_at=now_time,
+        )
         db.add(pending)
         return pending
 
+    @classmethod
+    def stage_admit(cls, db: Session, *, task_id: str, storage: str,
+                    src_path: str, state: str,
+                    now_time: str) -> Optional["TransferPending"]:
+        """
+        在调用方会话中暂存一条持久接纳记录。
+
+        相同存储与路径已存在时返回原记录，确保重复监控事件复用同一个任务标识。
+        :param db: 数据库会话
+        :param task_id: 任务标识
+        :param storage: 存储
+        :param src_path: 源文件路径
+        :param state: 持久状态
+        :param now_time: 当前时间
+        :return: 接纳记录
+        """
+        if not task_id or not storage or not src_path or not state:
+            return None
+        pending = db.execute(
+            select(cls).where(cls.storage == storage, cls.src_path == src_path)
+        ).scalars().first()
+        if pending:
+            return cast("TransferPending", pending)
+        pending = cls(
+            task_id=task_id,
+            storage=storage,
+            src_path=src_path,
+            state=state,
+            created_at=now_time,
+            updated_at=now_time,
+        )
+        db.add(pending)
+        return pending
+
+    @classmethod
+    def list_by_state(cls, db: Session, *, state: str,
+                      limit: Optional[int] = 5000) -> List["TransferPending"]:
+        """
+        按登记顺序列出指定持久状态的接纳记录。
+        :param db: 数据库会话
+        :param state: 持久状态
+        :param limit: 单次读取上限
+        :return: 接纳记录列表
+        """
+        if not state:
+            return []
+        return list(db.execute(
+            select(cls)
+            .where(cls.state == state)
+            .order_by(cls.created_at.asc(), cls.id.asc())
+            .limit(limit)
+        ).scalars().all())
+
+    @classmethod
+    def get_by_identity(cls, db: Session, *, storage: str,
+                        src_path: str) -> Optional["TransferPending"]:
+        """
+        按存储与源路径查询一条持久接纳记录。
+        :param db: 数据库会话
+        :param storage: 存储
+        :param src_path: 源文件路径
+        :return: 接纳记录
+        """
+        if not storage or not src_path:
+            return None
+        return cast(
+            Optional["TransferPending"],
+            db.execute(
+                select(cls).where(
+                    cls.storage == storage,
+                    cls.src_path == src_path,
+                )
+            ).scalars().first(),
+        )
+
+    @classmethod
+    def record_enqueue_failure(cls, db: Session, *, task_id: str,
+                               error: str, now_time: str) -> int:
+        """
+        在调用方会话中记录任务最近一次入队失败。
+        :param db: 数据库会话
+        :param task_id: 任务标识
+        :param error: 失败原因
+        :param now_time: 当前时间
+        :return: 更新的记录数
+        """
+        if not task_id:
+            return 0
+        return execute_dml(
+            db,
+            update(cls)
+            .where(cls.task_id == task_id)
+            .values(last_error=error, updated_at=now_time),
+            execution_options={"synchronize_session": False},
+        )
+
+    @classmethod
+    def discard_task(cls, db: Session, *, task_id: str) -> int:
+        """
+        在调用方会话中按任务标识删除接纳记录。
+        :param db: 数据库会话
+        :param task_id: 任务标识
+        :return: 删除的记录数
+        """
+        if not task_id:
+            return 0
+        return execute_dml(
+            db, delete(cls).where(cls.task_id == task_id),
+            execution_options={"synchronize_session": False},
+        )
+
     @classmethod
     def discard(cls, db: Session, storage: str, src_path: str) -> int:
         """

app/db/oper/transferpending.py

@@ -9,8 +9,9 @@ class TransferPendingOper(DbOper):
     """
     待整理文件登记管理。
 
-    只保存「存储 + 源文件路径」这一最小事实，用于在进程重启后把没走完整理链的
-    文件重新送回去，避免挂载故障重启后永久漏件。
+    保存稳定任务身份、存储、源文件路径和准入状态，用于在进程重启后把没走完
+    整理链的文件重新送回去，避免挂载故障重启后永久漏件。旧版路径登记接口继续
+    保留，供插件和兼容调用方使用。
     """
 
     def register(self, storage: str, src_path: str) -> Optional[TransferPending]:
@@ -30,6 +31,93 @@ class TransferPendingOper(DbOper):
             )
         )
 
+    def stage_admit(self, *, task_id: str, storage: str, src_path: str,
+                    state: str, now_time: str) -> Optional[TransferPending]:
+        """
+        在当前会话中暂存一条持久接纳记录。
+
+        适配器应传入显式 Session，使提交与回滚仍由应用用例对应的 UoW 管理。
+        :param task_id: 任务标识
+        :param storage: 存储
+        :param src_path: 源文件路径
+        :param state: 持久状态
+        :param now_time: 当前时间
+        :return: 接纳记录
+        """
+        return self._execute_sync_write(
+            lambda session: TransferPending.stage_admit(
+                session,
+                task_id=task_id,
+                storage=storage,
+                src_path=src_path,
+                state=state,
+                now_time=now_time,
+            )
+        )
+
+    def list_by_state(self, *, state: str,
+                      limit: Optional[int] = 5000) -> List[TransferPending]:
+        """
+        使用当前会话列出指定状态记录。
+        :param state: 持久状态
+        :param limit: 单次读取上限
+        :return: ORM 接纳记录列表
+        """
+        return self._execute_sync_query(
+            lambda session: TransferPending.list_by_state(
+                session,
+                state=state,
+                limit=limit,
+            )
+        ) or []
+
+    def get_by_identity(self, *, storage: str,
+                        src_path: str) -> Optional[TransferPending]:
+        """
+        使用当前会话按存储与源路径查询接纳记录。
+        :param storage: 存储
+        :param src_path: 源文件路径
+        :return: 接纳记录
+        """
+        return self._execute_sync_query(
+            lambda session: TransferPending.get_by_identity(
+                session,
+                storage=storage,
+                src_path=src_path,
+            )
+        )
+
+    def stage_record_enqueue_failure(self, *, task_id: str, error: str,
+                                     now_time: str) -> int:
+        """
+        在当前会话中暂存最近一次入队失败。
+        :param task_id: 任务标识
+        :param error: 失败原因
+        :param now_time: 当前时间
+        :return: 更新的记录数
+        """
+        return self._execute_sync_write(
+            lambda session: TransferPending.record_enqueue_failure(
+                session,
+                task_id=task_id,
+                error=error,
+                now_time=now_time,
+            )
+        )
+
+    def stage_discard_task(self, *, task_id: str) -> int:
+        """
+        在当前会话中暂存按任务标识删除接纳记录。
+        :param task_id: 任务标识
+        :return: 删除的记录数
+        """
+        return self._execute_sync_write(
+            lambda session: TransferPending.discard_task(
+                session,
+                task_id=task_id,
+            )
+        )
+
     def discard(self, storage: str, src_path: str) -> int:
         """
         注销一个待整理文件登记。