Cron one-shot 失火时的保底方案：链式任务加 watchdog 与手动恢复路径
问题/场景：社区报告 `schedule.kind=at` 在持续运行后可能卡死在 past due。前置条件：依赖 cron 链式 one-shot 任务。实施步骤：给链路加状态检查与 watchdog → 发现 stale 立刻告警并切换 `sessions_spawn` 恢复 → 清理陈旧任务。关键命令：openclaw gateway status。验证：watchdog 可在窗口内识别卡死并自动恢复。风险：这是临时缓解，根因修复版本仍需跟进。来源：Issue #20586。
GITHUBDiscovered 2026-02-19Author sauravbhattacharya001
Prerequisites
- You already run deferred jobs through cron one-shot (`schedule.kind: at`) in production-like loops.
- You can inspect cron state and are able to trigger fallback runs when a chain stalls.
Steps
- Persist expected next-run timestamp for each chain in a state file for external comparison.
- Run a periodic watchdog that flags jobs whose `nextRunAt` is in the past for too long.
- When stale is detected, send alert and trigger immediate fallback execution via `sessions_spawn`.
- Clean up stale one-shot jobs to avoid duplicated backlog and restart normal scheduling chain.
Commands
openclaw gateway statusVerify
If a one-shot chain stalls, watchdog alerts within one cycle and workload resumes through fallback path.
Caveats
- This is a mitigation pattern, not an upstream bug fix; keep tracking official patch progress.
- Aggressive watchdog intervals can create duplicate executions if idempotency is not designed（需验证）.
Source attribution
This tip is aggregated from community/public sources and preserved with attribution.
Open original source ↗