perf(test): parallelize unit test files to halve validate wall time

[kentcdodds]

Problem

npm run validate in CI takes ~5m20s, and the whole job ~6.5min. Per-check completion times inside a recent run (29892110417) show the unit test suite is the entire critical path:

Check	Completion
primitives, migrations	~2s
format, lint	~9s
typecheck	40s
MCP e2e	1m36s
Playwright e2e	3m15s
unit tests	5m20s

The vitest summary explains why: 327 files, 310s total, of which 180s is module import time and only 62s is actual test time — all serialized by fileParallelism: false in vitest-shared.ts (present since the original projects migration with no stated rationale).

Change

• Remove fileParallelism: false from the shared vitest project config, so node-unit (289 files) and workers-unit (37 files) run files across all cores. In CI, cap vitest at 3 workers so the concurrent Wrangler/Playwright startups are not starved on 4-core runners.
• Keep mcp-e2e explicitly serial with a comment: each file boots a real Wrangler dev server against the shared seeded e2e database.
• Raise the Playwright webServer startup budget to 180s in CI: with unit tests using all cores, the e2e server (client build + D1 migrations + Wrangler boot) exceeded the default 60s on the first CI attempt.
• Fix an nx cache-correctness gap found while measuring: worker:test / worker:test-mcp only hashed project files, so edits to the root vitest*.ts configs or tools/ (vite plugins + root-level test files the runner picks up) returned stale cache hits. Both targets now include those paths in their inputs.

Measurements

• Unit suite alone (4-core VM, cold nx cache): 196s → 82s.
• Full npm run validate locally with CI=true, cold cache: 1m54s, all 8 checks green.
• Real CI (run 29894981813, green): validate step 5m20s → 4m00s, unit tests 5m20s → 3m41s, total job ~6m35s → 5m27s. Playwright e2e is now the critical path.
• Stability: 6+ green runs of the parallel config across local, pre-push hooks, and CI.

Follow-up candidates (not in this PR)

• Playwright is the new critical path (workers: 1 due to documented SQLITE_BUSY contention on the shared local D1 file). Per-worker databases would unlock parallelism there.
• isolate: false for node-unit could cut the remaining per-file import cost further, but risks cross-file state leakage; not worth the flake risk without more investigation.

System recap — composes existing primitives (low risk)

Mode: recap · Base: main @ 36d84e2e · Head: e52169df

Classification: composes — no primitives added or changed. The diff touches only test-runner configuration (vitest-shared.ts, vitest.mcp-e2e.config.ts, playwright.config.ts) and nx cache inputs (nx.json); the primitives classifier matches zero primitives, the lowest-risk outcome. No runtime, storage, or surface code changes.

Primitives touched

None. All four changed files are repo tooling outside the primitives map.

  

Summary by CodeRabbit

• Tests
  • Improved test execution consistency by refining inputs used for cached test runs.
  • Limited test worker concurrency on CI for more predictable performance.
  • Updated end-to-end test execution to avoid parallel file startup when sharing dev servers and seeded data.
  • Increased the web server startup timeout (with CI-aware values) to reduce flaky startup failures.

[coderabbitai]

📝 Walkthrough

Walkthrough

Nx test targets now track Vitest and tooling inputs. CI worker limits, MCP e2e file sequencing, and Playwright web-server startup timeouts are updated to coordinate test execution and startup.

Changes

Test configuration

Layer / File(s)	Summary
Nx test input tracking nx.json	The test and test-mcp targets now include default, workspace Vitest configuration, and tooling inputs while retaining caching.
Test runtime coordination vitest-shared.ts, vitest.mcp-e2e.config.ts, playwright.config.ts	CI limits shared Vitest workers to three, MCP e2e tests disable file parallelism, and Playwright uses CI-aware web-server startup timeouts.

Estimated code review effort: 2 (Simple) | ~10 minutes

Possibly related PRs

• kentcdodds/kody#89: Updates Nx test target caching and input configuration.
• kentcdodds/kody#176: Modifies the MCP e2e Vitest configuration.
• kentcdodds/kody#721: Modifies the shared Vitest configuration.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly matches the main change: parallelizing unit test files to reduce validation time.

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch cursor/faster-validate-911e

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[github-actions]

🔎 Preview deployed: https://kody-pr-849.kody-a99.workers.dev

Worker: kody-pr-849
D1: kody-pr-849-db
KV: kody-pr-849-oauth-kv

Mocks:

• cloudflare: https://kody-pr-849-mock-cloudflare.kody-a99.workers.dev/__mocks (kody-pr-849-mock-cloudflare)