Skip to content

ci(cockpit-e2e): widen the fleet from 5 to 12 parallel legs - #1597

Open
blove wants to merge 1 commit into
mainfrom
blove/ci-cd-optimization-4d1e44
Open

blove wants to merge 1 commit into
mainfrom
blove/ci-cd-optimization-4d1e44

Conversation

@blove

@blove blove commented Nov 15, 2026

Copy link
Copy Markdown
Contributor

What

cockpit-e2e strategy.max-parallel 5 → 12, and the guard assertion on that value in scripts/ci-workflow.spec.mjs. No job, step, test, gate, needs edge, cache or permission changes.

Why

The cap was chosen when the fleet had 12 legs. It now has 82 at about 2.5 minutes each, so 5 wide has a floor near 40 minutes. In ten sampled runs the fleet took 42–87 minutes on full-fleet runs and was the last job to finish; CI — required and Deploy → Vercel both wait on it. The organization plan allows 20 concurrent hosted jobs, and the sampled runs peaked at exactly 20.

Expected effect (simulated, not yet measured)

Replaying recorded job durations at width 12 ends a full-fleet run about 20–26 minutes sooner with unchanged compute. That is a best case for a run with the pool to itself. With several runs overlapping the gain shrinks and non-fleet jobs finish later, though each run as a whole still ends sooner in every scenario replayed.

This PR is the measurement: it changes ci.yml, so it runs the full fleet with the same-run dependency artifact, comparable to recent merge-queue and main runs.

Verified locally

  • node --test scripts/ci-workflow.spec.mjs: 201 pass (fails 5 !== 12 when only the spec is changed)
  • scripts/cockpit-matrix.spec.mjs 48 pass, scripts/cockpit-dependencies.spec.mjs 37 pass
  • The Aggregate matrix outcome and Verify scoped CI jobs scripts, run with failed / cancelled / skipped / empty / malformed fleet results, deny each one

Not verified locally: scripts/ci-scope.spec.mjs did not finish on my machine; the diff does not touch the classifier.

Risks

  • Other jobs from the same run are often still queued when the fleet is dispatched, so a wider fleet competes with them.
  • A cold Playwright cache with 12 simultaneous downloads is unobserved.

Rollback

Revert this commit.

🤖 Generated with Claude Code

The fleet has grown from 12 legs to 82 while the cap stayed at 5, so it
alone takes 42-87 minutes and is the last job to finish on full runs; the
production deploy waits on it. 12 of the plan's 20 hosted slots leaves 8
for the rest of the run and sibling workflows.
@vercel

vercel Bot commented Nov 15, 2026 •

Copy link
Copy Markdown

The latest updates on your projects. Learn more about Vercel for GitHub.

Project Deployment Actions Updated
threadplane Ready Ready Preview Nov 15, 2026 4:35am UTC

Request Review

@blove
blove marked this pull request as ready for review November 16, 2026 00:26
@blove
blove added this pull request to the merge queue Nov 16, 2026
@github-merge-queue
github-merge-queue Bot removed this pull request from the merge queue due to failed status checks Nov 16, 2026

This branch was successfully deployed

1 active deployment
Preview – threadplane — 58efdee3 Deployed Nov 15, 2026 by vercel[bot]
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