Repository navigation
ci(lecture-validation): per-repo timeout (180 min for lecture-python.myst); log the runner CPU - #197
Conversation
…n.myst; log the runner CPU lecture-python.myst's CPU HTML build takes 72-118 min depending on the runner, so against one 120-minute timeout for every leg its result depended on which host it landed on: cancelled at 120 min on most weekly runs since 2026-08-03 (#141, #159). Matrix entries can now set timeout-minutes, defaulting to 120, and lecture-python.myst sets 180. Its runtime itself is #196. Each leg also records the runner's CPU model and core count, in the log and the job summary. Identical inputs have taken 71.9 and 116.5 min on different runners, and without the CPU a slow leg cannot be told from a slower build. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Gmq1LGR87Md5rq4RJshmMp
There was a problem hiding this comment.
Copilot review overview
🟢 Approval recommended
The changes are small, additive, and degrade gracefully to the current behavior even in the worst case, and the timeout expression and CPU-logging step are verified correct.
Review effort: Balanced
Findings: None
What changed in this PR
This PR adjusts the weekly container-validation workflow (test-containers-lectures.yml) to address the two root causes of the recurring red sweeps tracked in #159/#141: the shared 120-minute job ceiling that lecture-python.myst regularly exceeds, and the inability to distinguish a slow runner from a slow build. It makes the timeout configurable per matrix repo and adds per-leg CPU diagnostics, without attempting the separate upstream lecture fixes.
Changes:
- Make the job
timeout-minutesper-repo viamatrix.repo.timeout-minutes, defaulting to 120 and grantinglecture-python.myst180. - Add a "Record runner CPU" step that logs the CPU model and core count to the run log and the job summary.
- Document the change in
CHANGELOG.md.
| File | Description |
|---|---|
.github/workflows/test-containers-lectures.yml |
Replaces the fixed 120-minute ceiling with a per-repo timeout (180 for lecture-python.myst) and adds a CPU-recording step. |
CHANGELOG.md |
Adds an Unreleased "Changed" entry describing the per-repo timeout and CPU logging. |
💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.
…-d9pyll-validation-timeout # Conflicts: # CHANGELOG.md
This is the workflow half of the #159 follow-ups (diagnosis).
test-containers-lectures.ymlcan now settimeout-minutes, and the default stays 120.lecture-python.mystgets 180. Its CPU HTML build alone takes 72–118 min depending on the runner, so at 120 whether a leg passed depended on the host it landed on. Legs were cancelled at 120 min on most weekly runs since 2026-08-03 (test-containers-lectures: lecture-python.myst doubled to the 120-minute ceiling #141). The lecture's runtime itself is discussed in lecture-python.myst: CPU build time sits at the lecture-validation ceiling #196, and the workflow comment says to lower the limit once that lands.This does not fix
bayes_nonconj(numpyro 0.22.0). That fix belongs inlecture-python.myst, so this workflow stays red for that leg until it lands.Verification
matrixis an allowed context injobs.<id>.timeout-minutes, and entries without the key fall back to 120.sh. It printsRunner CPU: Intel(R) Xeon(R) Processor @ 2.10GHz (4 cores), and it falls back tounknownif/proc/cpuinfohas no model name.env:, not${{ }}insiderun:.lecture-python.mystwould still fail onbayes_nonconj, which would file another alert on 🔴 CI failure: Test Container Builds #159. The first real run is the next container build after merge.🤖 Generated with Claude Code
https://claude.ai/code/session_01Gmq1LGR87Md5rq4RJshmMp
Generated by Claude Code