fix: 必剪转写实例复用导致的上传状态残留与并发竞争

backend/app/transcriber/bcut.py

@@ -1,5 +1,6 @@
 import json
 import logging
+import threading
 import time
 from typing import Optional, List, Dict, Union
 
@@ -40,6 +41,10 @@ def __init__(self):
         self.session = requests.Session()
         self.task_id = None
         self.__etags = []
+        # 实例被 transcriber_provider 缓存复用，并发任务会交错读写上传会话状态
+        # （etags/upload_id/task_id），必须整段串行化
+        # ponytail: 全局实例锁，并发转写会排队；如需吞吐量应改为每任务独立实例
+        self._lock = threading.Lock()
 
         self.__in_boss_key: Optional[str] = None
         self.__resource_id: Optional[str] = None
@@ -59,6 +64,15 @@ def _load_file(self, file_path: str) -> bytes:
 
     def _upload(self, file_path: str) -> None:
         """申请上传"""
+        # 实例被 transcriber_provider 缓存复用，必须清掉上一次上传的会话状态，
+        # 否则 __etags 跨上传残留，提交的 etag 数与本次分片数不符，B 站会拒绝合并
+        self.__etags = []
+        self.__in_boss_key = None
+        self.__resource_id = None
+        self.__upload_id = None
+        self.__upload_urls = []
+        self.__download_url = None
+
         file_binary = self._load_file(file_path)
         if not file_binary:
             raise ValueError("无法读取文件数据")
@@ -169,6 +183,10 @@ def _query_result(self) -> dict:
     @timeit
     def transcript(self, file_path: str) -> TranscriptResult:
         """执行识别过程，符合 Transcriber 接口"""
+        with self._lock:
+            return self._transcript_locked(file_path)
+
+    def _transcript_locked(self, file_path: str) -> TranscriptResult:
         try:
             logger.info(f"开始处理文件: {file_path}")
 

backend/tests/test_bcut_state_reset.py

@@ -0,0 +1,74 @@
+import json
+
+from app.transcriber import bcut as bcut_module
+from app.transcriber.bcut import BcutTranscriber
+
+
+class _FakeResp:
+    def __init__(self, payload=None, etag=""):
+        self._payload = payload or {}
+        self.headers = {"Etag": etag}
+        self.url = ""
+
+    def raise_for_status(self):
+        pass
+
+    def json(self):
+        return self._payload
+
+
+class _FakeSession:
+    """模拟必剪上传三步接口：申请上传 / PUT 分片 / 提交合并"""
+
+    def __init__(self, per_size, upload_urls_count):
+        self.per_size = per_size
+        self.upload_urls_count = upload_urls_count
+        self.commit_payloads = []
+
+    def post(self, url, data=None, headers=None):
+        if url == bcut_module.API_REQ_UPLOAD:
+            body = json.loads(data)
+            return _FakeResp({
+                "data": {
+                    "in_boss_key": "boss",
+                    "resource_id": "res",
+                    "upload_id": "up",
+                    "upload_urls": [f"http://fake/{i}" for i in range(self.upload_urls_count)],
+                    "per_size": self.per_size,
+                    "size": body["size"],
+                }
+            })
+        if url == bcut_module.API_COMMIT_UPLOAD:
+            self.commit_payloads.append(json.loads(data))
+            return _FakeResp({"code": 0, "data": {"download_url": "http://fake/dl"}})
+        raise AssertionError(f"unexpected post: {url}")
+
+    def put(self, url, data=None, headers=None):
+        return _FakeResp(etag=f"etag-{url[-1]}")
+
+
+def _upload_file(transcriber, tmp_path, name, content, chunks):
+    f = tmp_path / name
+    f.write_bytes(content)
+    transcriber.session = _FakeSession(per_size=5, upload_urls_count=chunks)
+    transcriber._upload(str(f))
+    return transcriber.session.commit_payloads
+
+
+def test_second_upload_commits_only_its_own_etags(tmp_path):
+    t = BcutTranscriber()
+
+    # 第一次：3 字节，1 分片
+    commits = _upload_file(t, tmp_path, "a.mp3", b"aaa", chunks=1)
+    assert len(commits[0]["Etags"].split(",")) == 1
+
+    # 第二次（同一实例）：8 字节，2 分片——修复前会提交 3 个 etag
+    commits = _upload_file(t, tmp_path, "b.mp3", b"aaaaaaaa", chunks=2)
+    assert len(commits[0]["Etags"].split(",")) == 2
+
+
+def test_first_upload_unchanged(tmp_path):
+    t = BcutTranscriber()
+    commits = _upload_file(t, tmp_path, "a.mp3", b"aaa", chunks=1)
+    assert commits[0]["Etags"] == "etag-0"
+    assert commits[0]["UploadId"] == "up"