宜博 / 2080tiserver

Commit af31f0a885999c0cdf677028c75fdec886708394

Authored by yibo
1 parent 25a489dc

h3-console: 加批量取消/删除接口 

清理过载事故留下的重复排队任务时发现，取消/删除都只能按单条 id 调用，
脚本化清理只能一个个来。新增 POST /v1/videos/batch-cancel 和
/batch-delete，body 传 {"ids":[...]} 或 {status,kind,q} 筛选条件二选一；
筛选条件命中数超过 300 直接 400，避免条件写宽了误伤一大片。单条 cancel/
delete 的核心逻辑抽成 _cancel_job/_delete_job 供两边复用，批量场景里
一条失败（比如已经是终态）不影响其它条，成功/失败分别列出来。

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GFw9ZBh3eVjssk6quYQ4Ki
Inline Side-by-side
Showing 1 changed file with 86 additions and 5 deletions
h3-console/server.py
  View file @af31f0a
... ... @@ -396,8 +396,10 @@ async def create_video(req: dict):
396 396 return _row_view(row)
397 397  
398 398  
399   -@api.post("/v1/videos/{jid}/cancel")
400   -async def cancel_video(jid: str):
  399 +async def _cancel_job(jid: str) -> dict:
  400 + """真正干活的取消/停止逻辑。单条接口和下面的批量接口共用——批量场景需要一条失败不
  401 + 连累其它条，所以这里只管抛 HTTPException，由调用方决定是直接透传（单条）还是
  402 + catch 下来记进 failed 列表（批量）。"""
401 403 with db() as con:
402 404 row = con.execute("SELECT * FROM jobs WHERE id=?", (jid,)).fetchone()
403 405 if row is None:
... ... @@ -434,6 +436,11 @@ async def cancel_video(jid: str):
434 436 return _row_view(row)
435 437  
436 438  
  439 +@api.post("/v1/videos/{jid}/cancel")
  440 +async def cancel_video(jid: str):
  441 + return await _cancel_job(jid)
  442 +
  443 +
437 444 @api.post("/v1/videos/{jid}/prioritize")
438 445 async def prioritize_video(jid: str):
439 446 with db() as con:
... ... @@ -588,9 +595,8 @@ async def list_videos(limit: int = 20, offset: int = 0, trash: bool = False, q:
588 595 return {"jobs": [_row_view(r) for r in rows], "total": total}
589 596  
590 597  
591   -@api.delete("/v1/videos/{jid}")
592   -async def delete_video(jid: str):
593   - """软删除：只挪出主列表，不碰 gateway/api.py 上的任务本身（可能还在跑），不删磁盘文件。"""
  598 +def _delete_job(jid: str) -> dict:
  599 + """真正干活的软删除逻辑，单条/批量共用，理由同 _cancel_job。"""
594 600 with db() as con:
595 601 cur = con.execute("UPDATE jobs SET deleted_at=? WHERE id=? AND deleted_at IS NULL", (time.time(), jid))
596 602 if cur.rowcount == 0:
... ... @@ -598,6 +604,12 @@ async def delete_video(jid: str):
598 604 return {"id": jid, "deleted": True}
599 605  
600 606  
  607 +@api.delete("/v1/videos/{jid}")
  608 +async def delete_video(jid: str):
  609 + """软删除：只挪出主列表，不碰 gateway/api.py 上的任务本身（可能还在跑），不删磁盘文件。"""
  610 + return _delete_job(jid)
  611 +
  612 +
601 613 @api.post("/v1/videos/{jid}/restore")
602 614 async def restore_video(jid: str):
603 615 with db() as con:
... ... @@ -607,6 +619,75 @@ async def restore_video(jid: str):
607 619 return {"id": jid, "deleted": False}
608 620  
609 621  
  622 +# 2026-09-05 加：一次过载事故后要清理一批内容重复的排队任务，才发现只有按单条 id 操作的
  623 +# cancel/delete，脚本化清理只能一个个调，慢且容易漏。下面两个批量接口给 ids 列表和"筛选
  624 +# 条件"两种入参二选一——ids 优先且没有数量上限（调用方既然把 id 一个个列出来了，就是明确
  625 +# 知道要动哪些）；筛选条件复用跟 GET /v1/videos 一样的 status/kind/q 语义，但设了
  626 +# _BATCH_MAX_BY_FILTER 上限，防止筛选条件写宽了误伤一大片跟预期不符的任务——命中数超过
  627 +# 上限直接 400，让调用方缩小范围或换成显式 ids，而不是静默截断一部分。
  628 +_BATCH_MAX_BY_FILTER = 300
  629 +
  630 +
  631 +def _resolve_batch_targets(body: dict) -> list[str]:
  632 + ids = body.get("ids")
  633 + if ids:
  634 + return list(dict.fromkeys(ids)) # 去重但保留顺序，防止调用方列表里手滑写重了
  635 + status, kind, q = body.get("status") or "", body.get("kind") or "", body.get("q") or ""
  636 + if not (status or kind or q):
  637 + raise HTTPException(400, "必须提供 ids 列表，或至少一个筛选条件（status/kind/q），避免误操作全部任务")
  638 + where = ["deleted_at IS NULL"]
  639 + params: list = []
  640 + if q:
  641 + where.append("(id LIKE ? OR req_json LIKE ?)")
  642 + params += [f"%{q}%", f"%{q}%"]
  643 + if status:
  644 + where.append("status = ?")
  645 + params.append(status)
  646 + if kind:
  647 + where.append("partition = ?")
  648 + params.append(kind)
  649 + with db() as con:
  650 + total = con.execute(f"SELECT COUNT(*) FROM jobs WHERE {' AND '.join(where)}", params).fetchone()[0]
  651 + if total > _BATCH_MAX_BY_FILTER:
  652 + raise HTTPException(
  653 + 400,
  654 + f"筛选条件命中 {total} 条，超过单次批量操作上限 {_BATCH_MAX_BY_FILTER}，"
  655 + "请缩小筛选范围（加 status/kind/q）或改成显式 ids 列表",
  656 + )
  657 + rows = con.execute(f"SELECT id FROM jobs WHERE {' AND '.join(where)}", params).fetchall()
  658 + return [r["id"] for r in rows]
  659 +
  660 +
  661 +@api.post("/v1/videos/batch-cancel")
  662 +async def batch_cancel_videos(body: dict):
  663 + """批量取消/停止。body 传 {"ids": [...]} 或 {"status": "queued", "kind": "...", "q": "..."}
  664 + 筛选条件（至少给一个）。逐条按现有单条 cancel 语义处理，某一条已经是终态之类的失败不影响
  665 + 其它条，成功/失败分别列出来。"""
  666 + ids = _resolve_batch_targets(body)
  667 + succeeded, failed = [], []
  668 + for jid in ids:
  669 + try:
  670 + succeeded.append((await _cancel_job(jid))["id"])
  671 + except HTTPException as e:
  672 + failed.append({"id": jid, "error": e.detail})
  673 + return {"requested": len(ids), "succeeded": succeeded, "failed": failed}
  674 +
  675 +
  676 +@api.post("/v1/videos/batch-delete")
  677 +async def batch_delete_videos(body: dict):
  678 + """批量软删除（移入回收站），语义同单条 DELETE /v1/videos/{jid}——只是把任务从列表里挪走，
  679 + 不取消底层计算；真要连底层任务一起停掉，配合上面的 batch-cancel 一起调（谁先谁后不影响
  680 + 结果，两个操作互不冲突）。"""
  681 + ids = _resolve_batch_targets(body)
  682 + succeeded, failed = [], []
  683 + for jid in ids:
  684 + try:
  685 + succeeded.append(_delete_job(jid)["id"])
  686 + except HTTPException as e:
  687 + failed.append({"id": jid, "error": e.detail})
  688 + return {"requested": len(ids), "succeeded": succeeded, "failed": failed}
  689 +
  690 +
610 691 @api.get("/v1/backends")
611 692 async def backends():
612 693 resp = await client.get(f"{GATEWAY_URL}/health")
... ...