masknull/dsh-llm-resilience ↗★ 0
dsh-llm-resilience
A configuration card in the DSH Web UI for tuning LLM request retry policy and stream idle timeout (retry mode, max retries, retryable codes, backoff, jitter, timeout).
AI Analysis
核心用途是自定义 LLM 接口的重试次数、退避时间及超时策略，提升网络波动时的请求成功率。适合对接口稳定性有高要求的生产环境。
Install
$
npx -p @deepseek-ai/dsh dsh plugin --profile web add github:masknull/dsh-llm-resilienceREADME
Read the full README ↗Configuration
All fields are optional; blank = provider default.
| Field | Default | Meaning |
|---|---|---|
| Retry mode | normal | normal retries only eligible failure codes; always retries every failure. |
| Max retries | 5 | Eligible retries after the first request (normal mode). |
| Retryable codes | EMPTY_RESPONSE,RATE_LIMIT,SERVER,TIMEOUT,TRANSPORT | Comma-separated failure codes eligible for retry (normal mode). |
| Initial backoff (ms) | 500 | First local backoff delay before retrying. |
| Max backoff (ms) | 10000 | Ceiling for scheduled or provider-requested retry delay. |
| Jitter ratio | 0.1 | Symmetric random multiplier (0..1) around each backoff delay. |
| Stream idle timeout (ms) | 300000 | Max provider idle time while one stream read is outstanding. |