LLM completion error: context deadline exceeded

[rockdaboot]

This happens even for smaller reviews with a (slow) local model (ocr review --commit ...).

The review command has a --timeout value, help text says "concurrent task timeout in minutes (default 10)". But the context deadline seems to be ~1 minute.

How can I tune this, didn't find anything in the README, nor in the --help texts.

[lizhengfeng101]

Thanks for reporting this! The issue is confirmed.

The root cause is that certain internal sub-tasks (such as REVIEW_FILTER_TASK and RE_LOCATION_TASK) have hardcoded 60-second timeouts that override the user-provided --timeout flag. For slow local models, 60 seconds is often not enough.

We plan to add a dedicated model request timeout parameter to address this, so users can configure the per-request timeout independently for slower models.

Thanks again for pointing this out!

[ChrisYANG1111]

这个修复了吗？没有的话，请问什么时候修复呢？

[lizhengfeng101]

抱歉，暂时还没有修复。请你可以查看下是否有 issue 指出该问题，如果没有的话欢迎你创建一个 issue，便于开发者认领。

[ChrisYANG1111]

已创建

[guqihao-7]

Hi, I’d like to take this.

I understand the issue as --timeout being the per-file review timeout, while internal tasks like REVIEW_FILTER_TASK and RE_LOCATION_TASK still have their own 60s task-level LLM request timeouts. I’d like to add a dedicated --llm-timeout or --task-timeout option. By default it will preserve the existing task template timeouts, and only override them when the user explicitly sets --llm-timeout or --task-timeout.

Does this direction work for you?

hello，我可以尝试下吗？。

我的理解是：--timeout 是单个文件 review 的整体超时时间，而像 REVIEW_FILTER_TASK 和 RE_LOCATION_TASK 这样的内部任务仍然有自己的 60 秒 LLM 请求 timeout（在 task_template.json 里配置的）。我想新增一个专门的 --llm-timeout 或名字是 --task-timeout 的参数。默认情况下（不传这个参数的时候）它会使用现有 task template 里的 timeout，只有当用户显式设置 --llm-timeout 或 --task-timeout 时，才会覆盖这些 timeout。
这个方向ok吗？还是有其他方案做法？

[lizhengfeng101]

Hi @rockdaboot, just a quick update — this has been fixed in v1.6.1. We increased the hardcoded timeout for internal sub-tasks (REVIEW_FILTER_TASK and RE_LOCATION_TASK) from 60s to 180s.

If you still run into timeout issues with slower local models, please let us know. We plan to introduce a dedicated parameter (like --llm-timeout) in a future release for explicit control over per-task LLM HTTP request timeouts.