feat(fam-edge): 队列健壮性 - ①失败重试间隔 retry_interval_sec=1h（last_fail_at 记录，配额类故障等恢复再试）；②max_retries 2→10；③入队前 OpenCV 文件校验（可解码/大小/元数据，失败标 invalid 不入队）+ mtime 稳定窗口防 rclone 半成品 + 处理前二次确认

fam-edge/src/fam_edge/video_queue.py

@@ -21,7 +21,7 @@ from typing import List, Optional
 from .logger import setup_logger
 from .config_loader import load_config
 from . import oracle_db
-from .video_processor import VideoProcessor
+from .video_processor import VideoProcessor, validate_video
 
 logger = setup_logger('fam-edge.video_queue')
 
@@ -44,7 +44,12 @@ class VideoQueue:
         # 模型超时放大倍数：消费时在模型原配置 timeout 上 ×N
         self.timeout_multiplier = float(vp.get('timeout_multiplier', 2.0))
         # 单视频失败最大重试次数（failed 且 retry_count>=max_retries 不再消费）
-        self.max_retries = int(vp.get('max_retries', 2))
+        self.max_retries = int(vp.get('max_retries', 10))
+        # 失败重试最小间隔（秒）：配额类瞬时故障（如 429）需等其恢复再重试，避免短时间重复打爆
+        self.retry_interval_sec = int(vp.get('retry_interval_sec', 3600))
+        # 文件校验：入队前用 OpenCV 确认可解码；mtime 稳定窗口防 rclone 写入中的半成品
+        self.file_validate = bool(vp.get('file_validate', True))
+        self.stable_window_sec = int(vp.get('stable_window_sec', 60))
 
         self._queue: "queue.Queue[int]" = queue.Queue()
         self._queued: set = set()          # 已在队中的 video_id（防重复入队）
@@ -75,6 +80,26 @@ class VideoQueue:
             fn = os.path.basename(path)
             row = self.db.get_video_by_filename(fn)
             if row is None:
+                # 新文件：先过 mtime 稳定窗口 + 可解码校验，通过才登记入队；失败标记 invalid
+                if self.file_validate:
+                    if time.time() - os.path.getmtime(path) < self.stable_window_sec:
+                        logger.info(f"文件仍在写入（mtime 未稳定），跳过本轮: {fn}")
+                        continue
+                    ok, verr, vmeta = validate_video(path)
+                    if not ok:
+                        vid = self.db.ensure_video(fn, path, camera_name=self.camera_name)
+                        self.db.set_video_file_status(vid, False, verr)
+                        self.db._conn.execute(
+                            "UPDATE videos SET status='invalid' WHERE id=?", (vid,))
+                        self.db._conn.commit()
+                        logger.warning(f"文件校验失败，标记 invalid 不入队: {fn} ({verr})")
+                        continue
+                    if vmeta:
+                        vid = self.db.ensure_video(fn, path, camera_name=self.camera_name)
+                        self.db.set_video_file_status(vid, True, '', vmeta)
+                        logger.info(f"登记新视频并入队: {fn} (id={vid}, meta={vmeta})")
+                        self._enqueue(vid)
+                        continue
                 vid = self.db.ensure_video(fn, path, camera_name=self.camera_name)
                 logger.info(f"登记新视频并入队: {fn} (id={vid})")
                 self._enqueue(vid)
@@ -91,10 +116,26 @@ class VideoQueue:
             logger.info(f"启动恢复入队 {len(rows)} 个待处理视频")
 
     def _retry_allowed(self, row) -> bool:
-        if row['status'] == 'pending':
+        """是否允许消费/重试该视频：pending 直接放行；failed 需未超重试上限且距上次失败够久；invalid 永不放行"""
+        status = row['status']
+        if status == 'pending':
             return True
-        # failed：未超重试上限才允许再次消费
-        return int(row['retry_count'] or 0) < self.max_retries
+        if status == 'invalid':
+            return False
+        # failed：未超重试上限，且距上次失败 >= retry_interval_sec（等配额类故障恢复）
+        if int(row['retry_count'] or 0) >= self.max_retries:
+            return False
+        last_fail = row['last_fail_at'] or row['updated_at'] or ''
+        if last_fail:
+            try:
+                from datetime import datetime
+                lt = datetime.strptime(last_fail[:19], '%Y-%m-%d %H:%M:%S')
+                elapsed = (datetime.now() - lt).total_seconds()
+                if elapsed < self.retry_interval_sec:
+                    return False
+            except ValueError:
+                pass
+        return True
 
     def _enqueue(self, video_id: int):
         with self._lock: