Skip to content

docs: record that v0.3.2 was verified, not just diagnosed - #8

Merged
fatherlinux merged 1 commit into
mainfrom
docs/record-the-measurement
Nov 4, 2026
Merged

fatherlinux merged 1 commit into
mainfrom
docs/record-the-measurement

Conversation

@fatherlinux

Copy link
Copy Markdown
Member

The v0.3.2 changelog entry explained the regression and the fix, then stopped at the two numbers that proved the problem. A reader could not tell whether the fix actually worked.

It now carries all three measurements from trentina's sdlc ci step on a GitHub-hosted 4-vCPU runner — fresh VMs with no layer cache, so they are directly comparable:

step vs sequential
v0.2.2 sequential, 2 CPUs 927s (15m27s) —
v0.3.0 lanes, 1 CPU each 1478s (24m38s) +59%
v0.3.2 lanes, 2 CPUs each 704s (11m44s) −24%

Plus the per-check breakdown, and the one row that looks like a defect and is not: the main lane is 38s slower than it was sequentially, because the two lanes jointly ask for all four vCPUs where the sequential job asked for two. That is the trade working as designed — main was never the critical path, so trading 38s there for 223s overall is the point.

Also records why a self-hosted number is quoted from inside a single run rather than as a before/after: on a runner that keeps its layer cache and the hook-cache volume, absolute step times swung 40s across three runs of the same commit (83s, 123s, 127s), driven by startup 22s→2s and pytest 82s→62s. The within-run lane saving was stable at 64s / 68s / 68s.

Docs only — no version bump, so this does not cut a release.

🤖 Generated with Claude Code

https://claude.ai/code/session_01R7Y1jesjfgxHeqtKJCvay6

The v0.3.2 entry explained the regression and the fix and then stopped at the
two numbers that proved the problem. Anyone reading it later could not tell
whether the fix worked. It now carries all three measurements, the per-check
breakdown behind them, and the one row that looks wrong and is not: the `main`
lane comes out 38s slower than it was sequentially, because two lanes jointly
ask for all four vCPUs where the sequential job asked for two. That is the
trade working -- `main` was never the critical path.

Also records why a self-hosted figure is quoted from inside a single run:
absolute step times there swing 40s between runs of the same commit, because
the layer cache and the hook-cache volume survive.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01R7Y1jesjfgxHeqtKJCvay6

@github-actions github-actions Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Gatehouse found no issues.

Full review.

1 finding(s) dropped: their evidence is not in the diff.

@fatherlinux
fatherlinux merged commit 8bbe173 into main Nov 4, 2026
10 checks passed
@fatherlinux
fatherlinux deleted the docs/record-the-measurement branch November 4, 2026 16:42
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant