補 PPT 視覺 QA stale recovery

services/ppt_vision_service.py

@@ -150,6 +150,7 @@ def _record_audit_run(run: Dict[str, Any]) -> Dict[str, Any]:
     global _LAST_AUDIT_RUN
     payload = dict(run)
     payload['updated_at'] = payload.get('updated_at') or _now_label()
+    payload['pid'] = payload.get('pid') or os.getpid()
     _LAST_AUDIT_RUN = payload
     _write_persisted_audit_run(payload)
     return payload
@@ -176,13 +177,40 @@ def _timestamp_age_seconds(value: str | None) -> float | None:
         return None
 
 
+def _pid_exists(pid: Any) -> bool:
+    try:
+        pid_int = int(pid or 0)
+    except Exception:
+        return False
+    if pid_int <= 0:
+        return False
+    try:
+        os.kill(pid_int, 0)
+        return True
+    except OSError:
+        return False
+
+
 def _is_recent_active_audit_run(run: Dict[str, Any] | None) -> bool:
     if not run or run.get('status') not in {'queued', 'running'}:
         return False
+    if run.get('pid') and not _pid_exists(run.get('pid')):
+        return False
     age = _timestamp_age_seconds(run.get('updated_at') or run.get('started_at') or run.get('queued_at'))
     return age is None or age < _ACTIVE_AUDIT_TTL_SECONDS
 
 
+def _mark_stale_audit_run(run: Dict[str, Any]) -> Dict[str, Any]:
+    payload = dict(run)
+    payload.update({
+        'ok': False,
+        'status': 'error',
+        'finished_at': payload.get('finished_at') or _now_label(),
+        'error': 'background worker no longer running; audit state marked stale',
+    })
+    return _record_audit_run(payload)
+
+
 def _is_vision_infra_error(error: str | None) -> bool:
     text = (error or '').lower()
     return any(marker in text for marker in (
@@ -214,6 +242,7 @@ def _public_audit_run_payload(run: Dict[str, Any] | None) -> Dict[str, Any] | No
         'started_at': run.get('started_at') or '',
         'finished_at': run.get('finished_at') or '',
         'updated_at': run.get('updated_at') or '',
+        'pid': run.get('pid') or None,
         'filenames': [
             os.path.basename(str(name))
             for name in (run.get('filenames') or [])
@@ -235,6 +264,8 @@ def _public_audit_run_payload(run: Dict[str, Any] | None) -> Dict[str, Any] | No
 def get_ppt_vision_audit_status() -> Dict[str, Any]:
     """Return the current/last background visual QA run without touching DB."""
     raw_run = _load_last_audit_run()
+    if raw_run and raw_run.get('status') in {'queued', 'running'} and not _is_recent_active_audit_run(raw_run):
+        raw_run = _mark_stale_audit_run(raw_run)
     running = _AUDIT_LOCK.locked() or _is_recent_active_audit_run(raw_run)
     last_run = _public_audit_run_payload(raw_run)
     if running: