Retry engines that raise UnmetDependency
- 主要語言
- Jupyter Notebook
- 星號
- 2.6k
- 分支
- 1k
- PR 合併指標
- 30 天內沒有已合併 PR
描述
I'm using ipyparallel in a somewhat hackneyed scheme as a GPU cluster - upon job submission, each engine checks to see whether there are available GPU resources for running the job, and reports back to the scheduler if not. This can be accomplished with `@depend`, if the Scheduler can be set to not blacklist engines that raise dependency errors. What I'm really looking for is a `resource_busy` error though.
Is it possible to disable engine blacklist upon raising dependency errors? Or, even better, to more directly tell the scheduler that an engine is busy?
貢獻指南
研究方向
Start by tracing the scheduler's handling of @depend dependency errors and engine blacklisting. Determine whether the existing behavior can represent a busy engine or whether a resource_busy error is needed; done means the scheduler preserves engines that temporarily lack resources without treating them as failed.
由索引模型根據 Issue 內容生成。
評估
- 技術堆疊
- jupyter, python
- 領域
- distributed-systems
- Issue 類型
- 功能
- 難度
- 5/5
- 預估耗時
- 一週以上
- 活躍度
- 停滯
- 描述清晰度
- 基本清楚
- 新手友好度
- 30/100