test(tui): status_write_is_atomic_under_concurrent_reads fails on Windows with NotFound (rename over a file being read)

未关闭
#545 0 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 1 天内回复

还没有人认领这个 Issue。

评估

难度
4/5
预计耗时
3-5 天
新手友好度
45/100
Issue 类型
缺陷
描述清晰度
基本清楚
活跃度
活跃
技术栈
rust
领域
cli, testing

调研方向

Start with crates/atlas-tui/src/runtime/manager/health_check.rs around the failing test and trace write_status_atomic to atlas_core::fs::write_atomic. Then inspect the .tui-status.json readers used by atlas dev --status and the MCP status tools to see how they handle NotFound. Done means the Windows race is addressed consistently for the test and production readers.

由索引模型根据 Issue 内容生成。

描述

area: tui os: windows priority: 3 low type: bug

What fails

atlas-tui runtime::manager::health_check::tests::status_write_is_atomic_under_concurrent_reads failed once on Windows with a NotFound from the reader:

FAIL [   5.473s] (2651/3095) atlas-tui runtime::manager::health_check::tests::status_write_is_atomic_under_concurrent_reads
  thread '<unnamed>' (968) panicked at crates\atlas-tui\src\runtime\manager\health_check.rs:653:66:
  called `Result::unwrap()` on an `Err` value: Os { code: 2, kind: NotFound, message: "The system cannot find the file specified." }
  thread 'runtime::manager::health_check::tests::status_write_is_atomic_under_concurrent_reads' panicked at health_check.rs:664:31:
  called `Result::unwrap()` on an `Err` value: Any { .. }

Evidence: run 44087317372 (a throwaway workflow_dispatch matrix, 2026-11-17), job 134053020665, windows-2022 (image 20261112.366.7), runner GitHub Actions 1000025870, Test step of commit 8019de2f-based branch; 1 failure in 10 jobs of that run (the other 9, 5 of them on windows-2025-vs2026, passed). Not a runner loss: the step ended normally with 3094 passed, 1 failed.

Why it fails

The test has one thread rewrite .tui-status.json 300 times with write_status_atomic (atlas_core::fs::write_atomic: temp file + sync_all + fs::rename over the target) while another thread loops std::fs::read_to_string(path).unwrap() and expects the old or the new content, "never a partial file". On Windows the rename over an existing file that a reader has just opened or is opening can make the target momentarily not resolvable, so the reader gets NotFound (os error 2), which the test's .unwrap() turns into a panic. The Unix rename(2) never has that window.

Why it matters beyond the test

If the cause is the rename, any Windows reader of .tui-status.json (the atlas dev --status / MCP status tools read it while the runner rewrites it) can see a transient NotFound and may report "atlas dev is not running" for an instant. Worth checking the readers' handling together with the fix.

Not done here

No fix attempted (asked to file it). Candidates: make the test retry on NotFound (what the fix of #484/#485 did for similar fsync-speed races) only if the readers in production already tolerate it, or make write_atomic robust on Windows (ReplaceFileW, or a retry of the open in the shared reader).

Found while investigating #520 (runner losses); unrelated to them.

主要语言
Rust
星标
3
派生
1
平均合并
9 小时 51 分钟
30 天内合并 PR
310

环境准备

从这里开始

  1. 先读完整个 Issue,再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库,在一个分支上完成修改。
  4. 提交 Pull Request,并在描述里引用这个 Issue 编号。

codeatlasdev/atlas 的其他 Issue

查看 codeatlasdev/atlas 的全部 Issue

相似的 Issue

更多 Rust Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。