fix: reset stale remote agent updates after sac-api restart (0.4.3)

Clear zombie running jobs on deploy and add admin API to cancel a stuck job.

Co-authored-by: Cursor <cursoragent@cursor.com>

backend/app/services/host_remote_actions.py

@@ -219,6 +219,59 @@ def run_agent_fallback_action(db: Session, host: Host) -> AgentUpdateFallbackRes
     return execute_agent_update_fallback(db, host)
 
 
+def clear_stale_running_remote_actions(
+    db: Session,
+    *,
+    reason: str = "Прервано перезапуском SAC (фоновый worker не завершил job)",
+    host_ids: list[int] | None = None,
+) -> list[str]:
+    """Сбросить зависшие running после restart sac-api (daemon thread умер, запись в БД осталась)."""
+    from sqlalchemy import select
+
+    stmt = select(Host).where(Host.agent_update_state == "running")
+    if host_ids is not None:
+        stmt = stmt.where(Host.id.in_(host_ids))
+    rows = list(db.scalars(stmt).all())
+    if not rows:
+        return []
+
+    finished = _utcnow().isoformat()
+    hostnames: list[str] = []
+    for host in rows:
+        started = (host.remote_action or {}).get("started_at")
+        payload = dict(host.remote_action or {})
+        payload.update(
+            {
+                "title": payload.get("title") or "Remote action",
+                "status": "failed",
+                "message": reason,
+                "ok": False,
+                "finished_at": finished,
+                "started_at": started,
+                "target": payload.get("target") or host.hostname,
+            }
+        )
+        host.remote_action = payload
+        host.agent_update_state = "failed"
+        host.agent_update_last_error = reason[:2000]
+        hostnames.append(host.hostname or str(host.id))
+
+    db.commit()
+    with _lock:
+        for host in rows:
+            _running.discard(int(host.id))
+    return hostnames
+
+
+def cancel_host_remote_action(db: Session, host: Host, *, reason: str | None = None) -> bool:
+    """Отменить зависший running job вручную (admin)."""
+    if host.agent_update_state != "running":
+        return False
+    msg = reason or "Отменено администратором SAC"
+    cleared = clear_stale_running_remote_actions(db, reason=msg, host_ids=[int(host.id)])
+    return bool(cleared)
+
+
 def get_remote_action_status(host: Host) -> dict[str, Any]:
     payload = dict(host.remote_action or {})
     if not payload: