feat(repair): asyncio.Semaphore 并发 + 断点续跑 + CLI 参数

--concurrency 默认 16，--reaggregate-all 强制全量重聚合。
Semaphore 限视频并发数，视频内四步串行。progress 文件
asyncio.Lock + os.replace 原子写入。熔断阈值 max(.env, concurrency*2)。

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

.env.example

@@ -41,7 +41,7 @@ REDIS_URL=redis://localhost:6379/0
 LLM_TIMEOUT=120
 LLM_MAX_RETRIES=3
 LLM_RETRY_BASE_DELAY=2.0
-LLM_CIRCUIT_BREAKER_THRESHOLD=5
+LLM_CIRCUIT_BREAKER_THRESHOLD=5  # 实际阈值 = max(此值, concurrency*2)
 LLM_CIRCUIT_BREAKER_COOLDOWN=60
 LLM_TTFT_TIMEOUT=30
 LLM_INTER_TOKEN_TIMEOUT=15

tools/repair_trees.py

@@ -125,7 +125,7 @@ def should_skip_video(vid: str, finished: set[str], *, reaggregate_all: bool) ->
 # ---------------------------------------------------------------------------
 
 
-def _build_clients():
+def _build_clients(concurrency: int = 16):
     """构建 GovernedLLMClient（LLM + VLM）。
 
     返回:
@@ -141,6 +141,7 @@ def _build_clients():
     telemetry = SQLiteTelemetryRecorder(str(PROJECT_ROOT / "logs" / "repair_telemetry.db"))
 
     breaker_threshold = int(os.getenv("LLM_CIRCUIT_BREAKER_THRESHOLD", "5"))
+    breaker_threshold = max(breaker_threshold, concurrency * 2)
     breaker_cooldown = int(os.getenv("LLM_CIRCUIT_BREAKER_COOLDOWN", "60"))
     timeout_s = float(os.getenv("LLM_TIMEOUT", "120"))
     max_retries = int(os.getenv("LLM_MAX_RETRIES", "3"))
@@ -323,52 +324,87 @@ async def _repair_one_video(
 
 
 async def main_async(args: argparse.Namespace) -> None:
-    """异步主流程：遍历所有视频，逐个修复。"""
+    """异步主流程：并发修复视频。"""
     videos_dir = Path(args.videos_dir)
     srt_dir = Path(args.srt_dir)
     questions_dir = Path(args.questions_dir)
+    concurrency = args.concurrency
+    reaggregate_all = args.reaggregate_all
 
     # 扫描所有视频
     vid_dirs = sorted(
         d for d in videos_dir.iterdir()
         if d.is_dir() and (d / "tree.json").exists()
     )
-    logger.info("发现 {} 个视频待修复", len(vid_dirs))
+    logger.info("发现 {} 个视频", len(vid_dirs))
+
+    # 加载 progress
+    progress_path = PROJECT_ROOT / "logs" / PROGRESS_FILE
+    finished = load_progress(progress_path)
+    if finished:
+        logger.info("已完成 {} 个视频（从 progress 文件加载）", len(finished))
 
     if args.dry_run:
         logger.info("=== DRY RUN 模式：仅检测不修复 ===")
 
     # 构建客户端（dry_run 模式不需要）
-    llm, vlm = (None, None) if args.dry_run else _build_clients()
+    llm, vlm = (None, None) if args.dry_run else _build_clients(concurrency)
 
-    # 逐视频修复
+    # 过滤跳过的视频
-    all_stats = []
+    pending = []
+    skipped_count = 0
+    for vid_dir in vid_dirs:
+        vid = vid_dir.name
+        if should_skip_video(vid, finished, reaggregate_all=reaggregate_all):
+            skipped_count += 1
+            continue
+        pending.append(vid_dir)
+
+    if skipped_count:
+        logger.info("跳过 {} 个已完成视频，待处理 {} 个", skipped_count, len(pending))
+
+    # 并发编排
+    sem = asyncio.Semaphore(concurrency)
+    progress_lock = asyncio.Lock()
+    all_stats: list[dict] = []
+    stats_lock = asyncio.Lock()
     start_time = time.time()
-    for idx, vid_dir in enumerate(vid_dirs):
+    completed = 0
+
+    async def _process(vid_dir: Path) -> None:
+        nonlocal completed
+        async with sem:
             vid = vid_dir.name
             tree_path = vid_dir / "tree.json"
-        frames_dir = vid_dir  # frame_path 已含 "frames/" 前缀，不再嵌套
+            frames_dir = vid_dir
 
-        logger.info(
+            logger.info("开始修复 {}", vid)
-            "[{}/{}] 开始修复 {}",
-            idx + 1, len(vid_dirs), vid,
-        )
 
             stats = await _repair_one_video(
                 vid, tree_path, frames_dir, srt_dir, questions_dir,
                 llm, vlm, dry_run=args.dry_run,
             )
-        all_stats.append(stats)
 
-        # 每 10 个视频汇总一次
+            async with stats_lock:
-        if (idx + 1) % 10 == 0:
+                all_stats.append(stats)
+                completed += 1
+
+            # 无 error 且非 dry_run 才记 finished
+            if stats["error"] is None and not args.dry_run:
+                await save_progress(progress_path, progress_lock, vid)
+
+            # 进度日志
+            if completed % 10 == 0:
                 elapsed = time.time() - start_time
-            rate = (idx + 1) / elapsed * 60
+                rate = completed / elapsed * 60 if elapsed > 0 else 0
                 logger.info(
                     "进度: {}/{}, 已用 {:.0f}s, 速率 {:.1f} 视频/分钟",
-                idx + 1, len(vid_dirs), elapsed, rate,
+                    completed, len(pending), elapsed, rate,
                 )
 
+    tasks = [asyncio.create_task(_process(vd)) for vd in pending]
+    await asyncio.gather(*tasks)
+
     # 最终汇总
     elapsed = time.time() - start_time
     total_issues = sum(s["issues_found"] for s in all_stats)
@@ -379,11 +415,13 @@ async def main_async(args: argparse.Namespace) -> None:
     logger.info("=" * 60)
     logger.info("修复完成")
     logger.info("  视频总数: {}", len(all_stats))
+    logger.info("  跳过数: {}", skipped_count)
     logger.info("  问题总数: {}", total_issues)
     logger.info("  L3 修复数: {}", total_repaired)
     logger.info("  事实注入数: {}", total_injected)
     logger.info("  失败数: {}", total_errors)
     logger.info("  总耗时: {:.0f}s", elapsed)
+    logger.info("  并发数: {}", concurrency)
     logger.info("=" * 60)
 
     if total_errors > 0:
@@ -416,6 +454,17 @@ def parse_args() -> argparse.Namespace:
         action="store_true",
         help="仅检测不修复，不调用 VLM/LLM",
     )
+    parser.add_argument(
+        "--concurrency",
+        type=int,
+        default=16,
+        help="并发修复视频数（默认: 16）",
+    )
+    parser.add_argument(
+        "--reaggregate-all",
+        action="store_true",
+        help="强制全量重聚合，忽略 progress 文件",
+    )
     return parser.parse_args()