A hot laptop no longer fails the journeys, and a run says when macOS held back its cores (#396) #398

Merged
cmoriarty merged 4 commits from claude/browser-stall into main 2026-11-05 12:33:38 -05:00
Owner

Why it stalled. macOS sometimes takes the performance cores away from the scheduler, so everything shares the four efficiency cores. kern.sched_recommended_cores drops from 255 to 15. A 12 s browser step took 109 s and failed 4 journeys. Which journeys fail only depends on which are running, so no test is at fault. Heat is the likely reason, though macOS gives none: spinners with no browser set it off, but the same spinners later ran 345 s without it.

What changes

  • A laptop gets four times CI's journey limits (120 s per test, 40 s per wait). CI keeps 30 s and 10 s.
  • tools/test.sh ends with the cores it had, when it was held back, pass or fail.
  • docs/testing.md has a section, "A hot laptop".
  • Two failures that are not heat: run.spec.ts:5 clicked a tab under its close button, and the console missed what landed before its stream opened, so the bar's time went stale.

Cost. A journey that is really broken takes 40 s per wait to fail on a laptop, not 10 s.

Verified

  • ./tools/test.sh is green: 458 backend tests and 39 journeys, in 23 s.
  • The new journey fails on the old build, reading 10s, and passes 5 of 5 with the fix.
  • One browser step, started as the cores went, passed all 39 in 53.5 s on 1.7 performance cores on average. That is one run, and milder than the 109 s stall.

Not verified. The 10 runs in a row that braid#396 asked for were not run, at the owner's word.

Closes #396

🤖 Generated with Claude Code

**Why it stalled.** macOS sometimes takes the performance cores away from the scheduler, so everything shares the four efficiency cores. `kern.sched_recommended_cores` drops from 255 to 15. A 12 s browser step took 109 s and failed 4 journeys. Which journeys fail only depends on which are running, so no test is at fault. Heat is the likely reason, though macOS gives none: spinners with no browser set it off, but the same spinners later ran 345 s without it. **What changes** - A laptop gets four times CI's journey limits (120 s per test, 40 s per wait). CI keeps 30 s and 10 s. - `tools/test.sh` ends with the cores it had, when it was held back, pass or fail. - `docs/testing.md` has a section, "A hot laptop". - Two failures that are not heat: `run.spec.ts:5` clicked a tab under its close button, and the console missed what landed before its stream opened, so the bar's time went stale. **Cost.** A journey that is really broken takes 40 s per wait to fail on a laptop, not 10 s. **Verified** - `./tools/test.sh` is green: 458 backend tests and 39 journeys, in 23 s. - The new journey fails on the old build, reading `10s`, and passes 5 of 5 with the fix. - One browser step, started as the cores went, passed all 39 in 53.5 s on 1.7 performance cores on average. That is one run, and milder than the 109 s stall. **Not verified.** The 10 runs in a row that braid#396 asked for were not run, at the owner's word. Closes #396 🤖 Generated with [Claude Code](https://claude.com/claude-code)
Whatever lands between the page's read of a run's detail and its stream opening arrives in the stream's snapshot, as state and with no event, so nothing read again what only the detail carries. The bar's time stayed as first read. A busy machine widened that gap from milliseconds to seconds, and operator.spec.ts:167 failed on it with the bar reading 10s. The new journey holds the stream until the events are in: it fails on the old build and passes with the fix.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The run's name is on a tab too, and a worker's server keeps every running run in the tab bar, so a narrow tab's close button could lie over its title and the click never landed. The test's assertions do not change.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
macOS sometimes takes a Mac's performance cores away from the scheduler for minutes at a time: kern.sched_recommended_cores drops from 255 to 15, and a browser step of 12 s took 109 s and failed 4 journeys, whichever were running. A laptop now gets four times CI's limits, and CI keeps its own. tools/test.sh ends with the cores it had when it was held back, so a red run says why. docs/testing.md has what is known and what is not.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
chore(openspec): archive journeys on a hot laptop (#396)
All checks were successful
test / suite (pull_request) Successful in 43s
abf2e2eb16
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Sign in to join this conversation.
No reviewers
No labels
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
cmoriarty/braid!398
No description provided.