Workflow: sp run refreshes the code snapshot before the lock; re-clean erases exposure tombstones; missing run-dir undetected
维护者通常 1 天内回复
评估
调研方向
Start by reading workflow/bin/sp around line 262 and the Snakemake lock behavior, then inspect workflow/scripts/clean_exposure.py around lines 58–80 and workflow/scripts/completeness.py around lines 405–406. Reproduce each reported failure and compare clean_exposure.py with the tombstone merge in clean_tile.py. Done means all three defects are fixed and covered by checks for repeated runs, repeated cleaning, and a missing run directory.
由索引模型根据 Issue 内容生成。
描述
Three workflow defects found while rewriting the workflow's comments (comment sweep, PR to follow). Each was confirmed against develop (f2ab8a56) by a second reader; none is fixed by the sweep, which changes no code.
1. sp run refreshes the campaign's code snapshot before Snakemake takes its lock — workflow/bin/sp:262. snapshot_code rsyncs the checkout into $STATE_DIR/code unconditionally, before either Snakemake invocation, including for sp run -n. A second sp run (or a dry run) against a live campaign therefore overwrites the code that its queued and running jobs re-read when they start, and only afterwards does Snakemake refuse on its lock. This defeats the snapshot's purpose (keeping mid-campaign edits away from running jobs). A fix probably belongs in bin/sp: take or test Snakemake's lock first, or snapshot into a fresh directory per launch.
2. Re-cleaning a reclaimed exposure erases its tombstone's stage records — workflow/scripts/clean_exposure.py:58–80. The record starts as {}, is filled only from the manifests currently on disk, and replaces the existing tombstone without reading it. On an exposure already reclaimed (its manifests gone) a second clean writes an empty record. clean_tile.py already merges into the prior tombstone; clean_exposure.py should do the same.
3. completeness.py cannot detect a missing run directory — workflow/scripts/completeness.py:405–406. With neither --run-dir nor SP_RUN, Path("") becomes ., so if not str(run_dir) never fires and the check runs against the current directory instead of raising its configuration error. Validate the raw setting before converting it to a Path.
Claude (Opus) and GPT-6.1 Sol on behalf of Cail.
- 主要语言
- Python
- 星标
- 18
- 派生
- 14
- 平均合并
- 2 天 13 小时
- 30 天内合并 PR
- 35
环境准备
- 提供 Dockerfile 或 Docker Compose 文件
- 有 Pull Request 模板
- 阅读贡献指南
从这里开始
- 先读完整个 Issue，再读项目的贡献指南。
- 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
- Fork 仓库，在一个分支上完成修改。
- 提交 Pull Request，并在描述里引用这个 Issue 编号。
CosmoStat/shapepipe 的其他 Issue
-
难度 1/5 1 小时以内 新手友好度 85/100
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 75/100
维护者通常 1 天内回复
-
难度 4/5 3-5 天 新手友好度 45/100
维护者通常 1 天内回复
-
难度 5/5 一周以上 新手友好度 18/100
CosmoStat/shapepipe#956 · 2 条评论 ·
维护者通常 1 天内回复
-
Keep exposure stores node-local in single-allocation runs; persist only per-exposure products可能已有人在做 关联的 PR 仍在进行中或已合并。 未关闭
难度 4/5 3-5 天 新手友好度 18/100
维护者通常 1 天内回复
查看 CosmoStat/shapepipe 的全部 Issue
相似的 Issue
-
bug
难度 2/5 1-3 小时 新手友好度 76/100
lugassawan/swe-workbench#799 ·
维护者通常 1 天内回复
-
[Bug]: CPU without Triton: slot-mapping fallback has a stale signature after #57169, V1 inference fails on the first forward step可能已有人在做 关联的 PR 仍在进行中或已合并。 未关闭
难度 2/5 1-3 小时 新手友好度 74/100
vllm-project/vllm#61131 ·
维护者通常 1 天内回复
-
Non-string form field names raise AttributeError when files are attached可能已有人在做 @nightcityblade 今天认领。 未关闭
难度 2/5 1-3 小时 新手友好度 82/100
-
难度 2/5 1-3 小时 新手友好度 78/100
kubernetes-client/python#2735 ·
维护者通常 1 天内回复
-
desktop help wanted test
难度 2/5 1-3 小时 新手友好度 78/100
维护者通常 1 天内回复