• Home
  • Features
  • Pricing
  • Docs
  • Announcements
  • Sign In

cosmodesi / desilike / 35274491517
71%
main: 19%

Build:
Build:
LAST BUILD BRANCH: refactor-jax
DEFAULT BRANCH: main
Ran 17 Sep 2026 09:20PM UTC
Jobs 1
Files 101
Run time 1min
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst

17 Sep 2026 09:01PM UTC coverage: 71.221%. First build
35274491517

push

github

adematti
ci: give desilike/theories its own runners, and drop pytest -n

The serial 3.13 leg died twice with the runner agent killed under it, at 45.3 and then 37.4
minutes -- the same point in the suite both times, a different elapsed time each. That rules out
a timeout, and `timeout-minutes` was 90 with the step at 35. It was memory: measured on the
theory tests, two workers on one runner peak at 23.7 GB against the 16 GB a runner has.

`pytest -n auto` was the cause and is removed. It assumed four cores; the runner reports
`created: 2/2 workers`, so it was worth 2x at best, and it bought that by putting a second
accumulating process on the same 16 GB. Resident memory climbs monotonically through a run --
not a leak, nothing is retained that the collector can see, but the heap only becomes reclaimable
at teardown, so the ceiling scales with how many theory tests one process has got through.
Trimming between tests was measured and recovers nothing, because the peak is reached during a
test, not between them.

So parallelism moves up a level: desilike/theories gets three legs of its own, one per heavy
file, and separate legs mean separate runners. That lowers the ceiling where `-n` raised it, and
it is safe by construction rather than by measurement -- each leg runs a strict subset of the
suite that passed single-process in one 16 GB runner before any of this, so none can need more
than that already did. Wall clock follows the slowest leg rather than their sum.

Coverage keeps working across the split: each leg writes its own COVERAGE_FILE and uploads its
own artifact, and the combine job takes them all.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

18318 of 25720 relevant lines covered (71.22%)

0.71 hits per line

Jobs
ID Job ID Ran Files Coverage
1 35274491517.1 17 Sep 2026 09:20PM UTC 101
71.22
GitHub Action Run
Source Files on build 35274491517
  • Tree
  • List 101
  • Changed 0
  • Source Changed 0
  • Coverage Changed 0
Coverage ∆ File Lines Relevant Covered Missed Hits/Line
  • Back to Repo
  • Github Actions Build #35274491517
  • cec2da33 on github
  • Next Build on refactor-jax (#35276813025)
  • Delete
STATUS · Troubleshooting · Open an Issue · Sales · Support · CAREERS · ENTERPRISE · START FREE TRIAL · SCHEDULE DEMO
ANNOUNCEMENTS · TWITTER · TOS & SLA · Supported CI Services · What's a CI service? · Automated Testing

© 2026 Coveralls, Inc