Files
zTools2/app/services/upload_service.py
zikai 09705a8843 feat: SFTP 上传的文件经磁盘扫描补录后可在文件浏览页展示
SFTP 上传的文件直接落盘到 upload_root(chroot 根),不经 HTTP 路径,无 DB 记录,
文件浏览页(读 uploaded_file 表)看不到。新增磁盘扫描机制:

- UploadService.scan_sftp_files(force=False):遍历 uploads/ 下所有文件(排除 .work/
  与 .part 残品),与 DB 已有 storage_path 比对,磁盘有 DB 无的补录(source="sftp",
  流式算 size + sha256)。
- 频率限制:模块级 _last_scan_time,两次扫描间隔 < 3s 则跳过返回 0(只读 DB 缓存),
  force=True 强制扫。
- file_admin_controller 的 list 接口返回前触发 scan_sftp_files(),文件浏览页打开/
  刷新即自动同步 SFTP 文件(受 3s 频率限制,不会每次翻页都扫磁盘)。
- UploadedFileDAO 加 list_storage_paths() 批量查已记录路径集合供比对。

SFTP 文件补录后可正常下载/删除(与 HTTP 上传文件一致)。
2026-07-22 01:25:41 +00:00

266 lines
10 KiB
Python
Raw Permalink Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

"""上传服务:流式落盘 + 原子改名 + 元数据入库 + sha256 去重。
两条上传路径HTTP 整文件 / 分片拼接)共享本类的「存储路径生成 + 落库提交 +
sha256 去重」逻辑避免重复实现。SFTP 服务器仅作为文件暂存通道chroot 到
upload_root不再有 HTTP 登记接口。
"""
from __future__ import annotations
import hashlib
import logging
import os
import time
import uuid
from collections.abc import Iterator
from datetime import datetime, timezone
from pathlib import Path
from fastapi import UploadFile
from ..config import get_settings
from ..dao.uploaded_file_dao import UploadedFileDAO
from ..models.uploaded_file import UploadedFile
from ..schemas.file import FileUploadResponse, UploadedFileOut
logger = logging.getLogger("zikai.upload")
# 磁盘扫描频率限制:两次扫描至少间隔此秒数,否则跳过(只返回 DB 缓存)。
_SCAN_MIN_INTERVAL = 3.0
_last_scan_time: float = 0.0
def hash_stream(chunks: Iterator[bytes]) -> tuple[int, str]:
"""对一段字节块序列流式计算 (size, sha256)。供多条上传路径复用。"""
h = hashlib.sha256()
size = 0
for chunk in chunks:
size += len(chunk)
h.update(chunk)
return size, h.hexdigest()
class UploadService:
def __init__(self, dao: UploadedFileDAO) -> None:
s = get_settings()
self.dao = dao
self.upload_root = s.resolved_upload_dir()
self.chunk_bytes = s.storage.chunk_bytes
self.hash_on_upload = s.storage.sha256_on_upload
# ---------------- 写入 ----------------
def stream_to_disk(
self, file: UploadFile, source: str, uploaded_by: str,
) -> FileUploadResponse:
"""HTTP 流式上传:先写 ``<final>.part``DB 行提交后再 ``os.replace`` 成正式名。
失败仅会留下可识别的 ``.part`` 文件,由 start.sh 启动时统一清理。
"""
rel_path, abs_path = self.make_storage_path(file.filename or "")
part_path = abs_path.with_name(abs_path.name + ".part")
size, digest = self._write_part(file, part_path)
entity = UploadedFile(
storage_path=str(rel_path),
original_filename=os.path.basename(file.filename or rel_path.name),
content_type=file.content_type or "",
size_bytes=size,
sha256=digest,
source=source,
uploaded_by=uploaded_by,
)
return self.commit_entity(entity, part_path, abs_path)
# ---------------- 查询 ----------------
def scan_sftp_files(self, force: bool = False) -> int:
"""扫描 uploads/ 目录,把磁盘上有但 DB 未记录的文件补录入库。
SFTP 上传的文件直接落盘到 upload_rootchroot 根),不经过 HTTP 路径,
因此没有 DB 记录。本方法遍历磁盘文件,与 DB 已有 storage_path 比对,
为缺失项创建记录source="sftp",计算 size + sha256
频率限制:两次扫描间隔 < _SCAN_MIN_INTERVAL3s则跳过force=True 强制扫。
返回本次新补录的条数(跳过时返回 0
"""
global _last_scan_time
now = time.monotonic()
if not force and (now - _last_scan_time) < _SCAN_MIN_INTERVAL:
return 0
_last_scan_time = now
known = self.dao.list_storage_paths()
new_count = 0
# 排除分片会话暂存目录与 .part 残品
skip_dirs = {".work"}
for abs_path in self.upload_root.rglob("*"):
if not abs_path.is_file():
continue
if abs_path.suffix == ".part":
continue
# 跳过 .work 目录下的任何文件
rel = abs_path.relative_to(self.upload_root)
if rel.parts and rel.parts[0] in skip_dirs:
continue
storage_path = str(rel).replace("\\", "/")
if storage_path in known:
continue
# 磁盘有、DB 无:补录
try:
size = abs_path.stat().st_size
sha256 = self._hash_file(abs_path)
except Exception as exc:
logger.warning("扫描文件失败 path=%s: %s", abs_path, exc)
continue
entity = UploadedFile(
storage_path=storage_path,
original_filename=abs_path.name,
content_type="",
size_bytes=size,
sha256=sha256,
source="sftp",
uploaded_by="sftp",
)
try:
self.dao.create(entity)
new_count += 1
known.add(storage_path)
except Exception as exc:
logger.warning("补录 SFTP 文件失败 path=%s: %s", storage_path, exc)
if new_count:
logger.info("SFTP 文件扫描补录 %d", new_count)
return new_count
def _hash_file(self, path: Path) -> str:
"""流式计算文件 sha256避免大文件一次性读入内存"""
h = hashlib.sha256()
with path.open("rb") as f:
while chunk := f.read(self.chunk_bytes):
h.update(chunk)
return h.hexdigest()
def list_files(self, limit: int = 100, offset: int = 0) -> tuple[int, list[UploadedFileOut]]:
total = self.dao.count()
rows = self.dao.list(limit=limit, offset=offset)
items = [UploadedFileOut.model_validate(r) for r in rows]
return total, items
def get_out(self, file_id: int) -> UploadedFileOut | None:
row = self.dao.get_by_id(file_id)
return UploadedFileOut.model_validate(row) if row else None
def find_by_sha256(self, sha256: str) -> UploadedFileOut | None:
row = self.dao.get_by_sha256(sha256)
return UploadedFileOut.model_validate(row) if row else None
def get_out_with_disk_path(self, file_id: int) -> tuple[UploadedFileOut | None, Path | None]:
"""合并查询:一次 DB 读取同时返回 (元数据, 磁盘绝对路径)。
供下载/删除路径复用,避免原先 get_out + resolve_disk_path 各查一次的重复读。
"""
row = self.dao.get_by_id(file_id)
if row is None:
return None, None
out = UploadedFileOut.model_validate(row)
path = (self.upload_root / row.storage_path).resolve()
return out, path
def delete_file(self, file_id: int) -> bool:
"""硬删除:删磁盘文件 + 删 DB 行。磁盘缺失不阻断 DB 清理。返回是否命中。
供单删/批删复用,保证删除语义一致。磁盘删除失败仅记日志,仍清 DB 行
(保证列表不再显示),避免磁盘文件泄漏却无任何记录。
"""
out, path = self.get_out_with_disk_path(file_id)
if out is None:
return False
if path is not None:
try:
Path(path).unlink(missing_ok=True)
except Exception as exc:
logger.warning("删除磁盘文件失败 file_id=%s path=%s: %s", file_id, path, exc)
self.dao.delete(file_id)
return True
# ---------------- 共享 helper供本类与 ChunkUploadService 复用) ----------------
@staticmethod
def _safe_ext(filename: str) -> str:
return os.path.splitext(os.path.basename(filename))[1]
@staticmethod
def to_response(row: UploadedFile, *, deduplicated: bool = False) -> FileUploadResponse:
return FileUploadResponse(
id=row.id,
filename=row.original_filename,
size_bytes=row.size_bytes,
sha256=row.sha256,
storage_path=row.storage_path,
uploaded_at=row.uploaded_at,
deduplicated=deduplicated,
)
def make_storage_path(self, original_filename: str) -> tuple[Path, Path]:
"""生成 ``(rel_path, abs_path)``abs_path 的父目录已创建。"""
now = datetime.now(timezone.utc)
rel_dir = Path(f"{now:%Y}/{now:%m}")
(self.upload_root / rel_dir).mkdir(parents=True, exist_ok=True)
rel_path = rel_dir / f"{uuid.uuid4().hex}{self._safe_ext(original_filename)}"
return rel_path, self.upload_root / rel_path
def commit_entity(
self, entity: UploadedFile, src_path: Path, dest_path: Path,
) -> FileUploadResponse:
"""落 DB 行后把 src_path 原子改名到 dest_path任一失败回滚已生成的副作用。"""
try:
saved = self.dao.create(entity)
except Exception:
src_path.unlink(missing_ok=True)
raise
try:
os.replace(src_path, dest_path)
except Exception:
try:
self.dao.delete(saved.id)
finally:
src_path.unlink(missing_ok=True)
raise
return self.to_response(saved)
def dedup_or_commit(
self, entity: UploadedFile, src_path: Path,
) -> FileUploadResponse:
"""公共尾部:按 entity.sha256 去重,命中则删 src 返回旧行;否则 commit。
供分片拼接等「先算出 sha256 再决定落盘」的路径复用,与 stream_to_disk
(边写边算、无独立 src的区别在于这里 sha256 已在 entity 上。
"""
existing = self.dao.get_by_sha256(entity.sha256)
if existing is not None:
src_path.unlink(missing_ok=True)
return self.to_response(existing, deduplicated=True)
rel_path, abs_path = self.make_storage_path(entity.original_filename)
entity.storage_path = str(rel_path)
return self.commit_entity(entity, src_path, abs_path)
# ---------------- 内部 ----------------
def _write_part(self, file: UploadFile, part_path: Path) -> tuple[int, str]:
"""把上传流写到 part_path 并 fsync返回 (size, sha256)。失败时清理残品。"""
hasher = hashlib.sha256() if self.hash_on_upload else None
size = 0
try:
with part_path.open("wb") as out:
while chunk := file.file.read(self.chunk_bytes):
out.write(chunk)
size += len(chunk)
if hasher is not None:
hasher.update(chunk)
out.flush()
os.fsync(out.fileno())
except Exception:
part_path.unlink(missing_ok=True)
raise
return size, (hasher.hexdigest() if hasher is not None else "")