openclaw - 💡(How to fix) Fix Model failover should handle silent API hangs (no response), not just error responses [1 participants]

When the primary model (e.g. Opus) accepts a request but never responds (silent hang), OpenClaw waits until the job timeout kills the request. The fallback chain (Opus → Sonnet → Qwen) does not engage because there's no error response to trigger failover — just silence.

TL;DR

Implement a timeout or a mechanism to detect silent hangs in the primary model to trigger the fallback chain.

Guidance

• Identify the maximum allowed response time for the primary model (e.g., Opus) and implement a timeout to detect silent hangs.
• Modify the OpenClaw configuration to treat a silent hang as an error, allowing the fallback chain to engage.
• Investigate the Opus API to determine why it accepts requests but returns zero tokens, potentially indicating an internal issue.
• Consider adding logging or monitoring to detect and report silent hangs, facilitating easier debugging and issue resolution.

Example

No code snippet is provided due to the lack of specific implementation details in the issue.

Notes

The solution may vary depending on the specific implementation of the OpenClaw and Opus APIs. Additional information about the APIs and their configurations might be necessary to provide a more detailed fix.

Recommendation

Apply a workaround by implementing a timeout or a detection mechanism for silent hangs, as this will allow the fallback chain to engage and provide a more robust system.