Skip to content

docs(ci): record split runtime measurements on main #2210

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
schlessera merged 1 commit into from
Oct 16, 2026
Merged

docs(ci): record split runtime measurements on main #2210

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
27 changes: 25 additions & 2 deletions docs/decisions/ci-utility.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -253,8 +253,31 @@ serial runtime category: 13.522/9.719/4.571 seconds for the probes, 424.345 seco
for hosted admission and 250.566 seconds for shared-process cleanup. Its four-CPU
VM sampled 2.540 GB used of 16.766 GB and 884.56 active CPU-seconds. These are
attributed historical observations, not a controlled current-source speedup.
Current two-job phase and full-job/setup measurements are required to qualify the
split against the longest single phase plus 90-second wall-time criterion.
The split's [runtime-only experiment 38920698136, attempt 1](https://github.com/schlessera/brain-kit/actions/runs/38920698136)
ran source `e7fea8b1ecb4a9f89f8651eca37e0bca797c20ad`, tree
`4c3f9f786a6838b02f43ae8753253c927956b2f8`. Actual checkout logs and
immutable checkout/resource artifacts agree. All five phases succeeded:
15.75 / 12.01 / 5.38 seconds for the probes, 265.72 seconds for shared-process
Chrome/cleanup and 427.17 seconds for separate hosted admission. Ordinary
runtime totaled 298.86 seconds (299-second proof step, 352-second whole job);
hosted admission took a 427-second step and 480-second whole job. The longest
whole job, including setup, satisfies the longest phase plus 90-second ceiling
(480 <= 517.17 seconds). This measurement workflow is an experiment, not a
production confirmation event.

[Automatic main run 38920963140, attempt 1](https://github.com/schlessera/brain-kit/actions/runs/38920963140)
independently executed merged source `80e028d84b7686d6cf864d8166029b1142de8756`,
tree `5f75530d5f191f6615f78736a7c67577b0cdbf84`, in both successful runtime
jobs. Its five phases succeeded in 11.90 / 8.85 / 3.93 / 194.25 / 424.80 seconds.
Ordinary runtime totaled 218.93 seconds; hosted admission's 425-second proof
step and 478-second whole job satisfy the same ceiling (478 <= 514.80 seconds).
The ordinary/hosted VM receipts record sampled peaks of 2.53 / 1.93 GB and
445.21 / 328.54 active CPU-seconds on separate 4-vCPU, 16.77-GB runners.
These source-bound main jobs establish phase coverage and the wall-time bound;
they do not establish a whole-workflow pass or a controlled speedup. The
experiment retains its own source and event identity. The separate
confirmation-event requirement remains tracked in
[#2128](https://github.com/schlessera/brain-kit/issues/2128).

A complete unit selection already includes cheap invariants, so CI does not run
that subset twice. No independent test/type job waits for packaging. The final
Expand Down
Loading