fix(ci): isolate HTTP end-to-end suites and stop the SCIM readiness flake

[waleedlatif1]

Summary

• Flake root cause: the SCIM step starts next dev and waits for /api/health, whose first request cold-compiles the app under Turbopack. Across 40 recent green runs readiness took 42–93s (step totals up to 209s), and today's staging push hit >120s — the fixed 120s deadline failed on the slow tail. The server log was never uploaded, so failures were undiagnosable.
• New End-to-end over real HTTP job holds SCIM + the four search HTTP suites, on its own Postgres (provisioned via db:migrate, the production path). Previously SCIM ran on both integration matrix legs, which are the PR critical path (~13 min avg vs 7.6 min Lint and Test); the search suites were gated by matrix.provision == 'push'. Isolation also matters: the SCIM app boots hosted, which starts background usage replay against DATABASE_URL.
• Readiness: 300s deadline (hang catcher only), step timeout-minutes: 12, fail-fast on server exit, server log tail printed inline, and the full scim-next.log uploaded with the reports (one artifact instead of five upload steps).
• setup-workspace composite action replaces four copies of Bun/Node/cache/install setup. Cache keys are unchanged (existing sticky disks are reused); the integration legs now also mount the lockfile-keyed node_modules disk.
• Removed a duplicated comment paragraph in the base-ref step.

Expected effect: ~2 min off each integration leg (the PR critical path), SCIM run once instead of twice, and the readiness flake gone.

ci-score (StarSling) rates the repo 91/100 before and after; its one failing check (fetch-depth: 0 in helm.yml) is a false positive — ct diffs against the PR base.

Test plan

• actionlint — no new findings
• CI on this PR: new End-to-end over real HTTP job green; integration legs no longer run SCIM

[vercel]

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

Project	Deployment	Actions	Updated
docs	Ready	Preview	Oct 1, 2026 9:33pm UTC

[cubic-dev-ai]

No issues found across 2 files

Confidence score: 5/5

• Automated review surfaced no issues in the provided summaries.
• No files require special attention.

_{Re-trigger cubic}

[greptile-apps]

Confidence Score: 4/5

[Medium risk] Refactors CI workflow setup into a reusable action.

The PR should not merge until the SCIM step has enough time for both a slow startup and the full suite.

Findings

1.  SCIM step can time out ▶
2.  SCIM CI guide is outdated ▶

Summary

The PR moves SCIM and four search HTTP suites into a dedicated, migration-provisioned job, extracts shared workspace setup into a composite action, and extends SCIM readiness while collecting failure logs. The SCIM step’s new timeout may truncate a slow suite, and the existing SCIM CI guide needs to reflect the move.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart LR
  CI[Test and Build workflow] --> Integration[PostgreSQL integration: push and migrate]
  CI --> HTTP[HTTP end-to-end: migrate]
  CI --> Lint[Lint and Test]
  CI --> Build[Build App]
  HTTP --> Search[Four search suites]
  Search --> SCIM[Readiness wait and SCIM suite]

Loading

_{Reviews (1) · Last reviewed commit: "fix(ci): isolate HTTP end-to-end suites ..."}