修复 rate-limit 冷却级联锁死（30s→60s→120s→5min）

解决“多个模型故障切换后统一进入超长冷却、机器人看起来完全失联”场景：PR #14574 将 rate-limit 退避改为温和阶梯，并在 gateway 启动时清理过时冷却。

GITHUBDiscovered 2026-02-12Author JamesEBall

Prerequisites

You run multiple provider/model profiles and have seen recurring rate-limit cooldown events.
You can inspect gateway logs and perform controlled restart tests.

Steps

Upgrade to a build including PR #14574 and restart gateway once to apply startup cooldown cleanup.
Simulate or replay rate-limit failures and confirm cooldown progression follows 30s → 60s → 120s → max 5min.
Restart gateway after transient quota recovery and verify stale non-billing cooldowns are cleared.
Keep billing-disabled profiles untouched and monitor for repeated true quota exhaustion.

Commands

openclaw gateway status

openclaw gateway restart

openclaw status

Verify

After transient rate limits, service recovers within minutes rather than entering hour-long lockout, and restart no longer preserves stale non-billing cooldowns.

Caveats

This change does not fix billing exhaustion; billing-disabled profiles should remain blocked.
If upstream provider limits remain saturated, retries still fail despite gentler backoff（需验证）.

Source attribution

This tip is aggregated from community/public sources and preserved with attribution.

Open original source ↗

Visit original post