Repository navigation
ci: hosted 子进程启动耗时证据 (#81/#144/#40) - #200
Conversation
A dispatch-only job times a hundred plain PowerShell 5.1 starts and fifty node --version starts on windows-2025, reporting percentiles and counts over the fixed fixture deadlines without changing any deadline.
Adds the dispatch job output as fixed results: 100 PowerShell 5.1 starts (p50 110ms, max 1.83s) and 50 node starts (p50 16ms, max 0.74s), none near the 10-12s fixture deadlines.
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 39b9038667
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| - 因此 `ps51` 在 CI 里出现 **10219 ms** 的 `timed_out`（#81）不能归因于启动成本；Node 夹具的 10 秒超时（#144）同理； | ||
| - 三处（ps51 10.2 s、纯 PowerShell 12 s、Node 10 s）都属于"纯启动/纯脚本、几乎无输出、刚好越过期限"，与本次测得的正常分布相差 1–2 个数量级，指向**宿主级停顿**（安全扫描、VM 调度、存储阻塞）。 |
There was a problem hiding this comment.
The new job measures only 100 PowerShell launches and 50 Node launches on an otherwise idle runner, so observing no 10-second launch does not exclude rare startup delays in the failing CI environment. In particular, scanning, scheduling and storage stalls can themselves occur during process startup; they are not mutually exclusive explanations. Treating startup cost as ruled out could prematurely narrow the investigations in #81/#144/#40. Report the observed idle-run distribution and retain the timeout explanation as unresolved unless measurements from the failing workload distinguish startup from subsequent execution; update the equivalent conclusion in validation.json as well.
AGENTS.md reference: AGENTS.md:L12-L13
Useful? React with 👍 / 👎.
内容
新增两项：
windows-2025-startup-latency：在 windows-2025 上测 100 次纯 PowerShell 5.1 启动与 50 次node --version，统计分位数与 ≥1s/≥5s/≥10s 计数（不改任何期限）；research/ui/results/2026-10-30-hosted-startup-latency/（README + validation.json）。结果（run 40935653206）
解释
纯启动成本比固定期限小 1–2 个数量级，因此 CI 里
ps51的 10219 ms（#81）与 Node 夹具的 10 秒超时（#144）不能归因于启动成本；与 #144 早先的 Node 计时（warm 63 ms、冷启动 1.47 s）互相印证，指向宿主级停顿。边界
Refs #81 #144 #40