ci: run PR unit shards on free hosted ARM as 12 shards (#34906)

* ci: run PR unit shards on free hosted ARM as 12 shards

Enterprise's 1,000-job limit makes more, smaller shards free. A 12-shard
cold run on ubuntu-24.04-arm measured 361s median / 412s max per shard,
against ~390s / ~441s for five warm Blacksmith 8-vCPU shards.

The Blacksmith-store module cache producer becomes dispatch-only since
hosted shards cannot read it.

* TEMP: force 12 shards on this PR (repo var is still 5); drop before merge

* test: size the full-inventory routing check from FULL_SHARD_COUNT

* Revert TEMP 12-shard override; the repository variable is now 12

---------

Co-authored-by: OrcaWin <293788423+OrcaWin@users.noreply.github.com>

.github/workflows/pr.yml

@@ -234,8 +234,8 @@ jobs:
           PREFLIGHT_PRIOR_SUCCESS: ${{ job.status == 'success' }}
           ORCA_UNIT_SELECTION_MODE: ${{ vars.ORCA_UNIT_SELECTION_MODE || 'shadow' }}
           ORCA_PULL_REQUEST_DRAFT: ${{ github.event.pull_request.draft }}
-          # Five shards use fewer runner minutes; the repository variable can request 2 to 10.
-          ORCA_UNIT_FULL_SHARD_COUNT: ${{ vars.ORCA_UNIT_FULL_SHARD_COUNT || '5' }}
+          # Twelve 4-vCPU hosted shards match the old 8-vCPU wall time; the repository variable can request 2 to 16.
+          ORCA_UNIT_FULL_SHARD_COUNT: ${{ vars.ORCA_UNIT_FULL_SHARD_COUNT || '12' }}
         run: |
           if [ "$PREFLIGHT_PHASE_SELECTED" != true ] || [ "$PREFLIGHT_PRIOR_SUCCESS" != true ]; then exit 0; fi
           node config/scripts/ci-unit-plan.mjs
@@ -896,7 +896,8 @@ jobs:
     uses: ./.github/workflows/unit-tests.yml
     with:
       node_versions: '["24"]'
-      runner: blacksmith-8vcpu-ubuntu-2404
+      # Free hosted ARM: 12 shards measured 361s median cold vs ~390s for 5 warm Blacksmith 8-vCPU shards.
+      runner: ubuntu-24.04-arm
       phase_diagnostics: true
       # Restore main's transform cache; the `ci:cold-unit-cache` label or ORCA_UNIT_MODULE_CACHE=off forces cold.
       module_cache: ${{ (vars.ORCA_UNIT_MODULE_CACHE == 'off' || contains(github.event.pull_request.labels.*.name, 'ci:cold-unit-cache')) && 'off' || 'warm' }}

.github/workflows/unit-module-cache.yml

@@ -1,29 +1,9 @@
 name: Unit transform cache producer
 
-# Cold full unit run on main that publishes per-shard Vitest transform caches for PRs.
+# Cold full unit run on main that publishes per-shard Vitest transform caches.
+# Dispatch only: PR shards run on hosted runners, which cannot read this Blacksmith-store cache.
 on:
-  schedule:
-    - cron: '17 */6 * * *'
   workflow_dispatch:
-  push:
-    branches: [main]
-    # Cache-key inputs (see ci-unit-module-cache.mjs) change the key; reseed right away.
-    paths:
-      - '.github/workflows/unit-module-cache.yml'
-      - '.github/workflows/unit-tests.yml'
-      - 'config/vitest.config.ts'
-      - 'config/scripts/vitest*'
-      - 'config/scripts/ci-unit-module-cache*.mjs'
-      - 'config/patches/**'
-      - 'config/scripts/workspace-source-exports.mjs'
-      - 'config/dark-build-switches.cjs'
-      - 'config/pnpmfile.cjs'
-      - '.npmrc'
-      - '**/package.json'
-      - '**/tsconfig*.json'
-      - 'pnpm-lock.yaml'
-      - 'pnpm-workspace.yaml'
-      - 'config/.bun-version'
 
 concurrency:
   group: unit-module-cache-${{ github.ref }}

.github/workflows/unit-plan.yml

@@ -36,7 +36,7 @@ jobs:
           ORCA_UNIT_SELECTION_MODE: ${{ inputs.selection_mode }}
           ORCA_PULL_REQUEST_DRAFT: ${{ github.event.pull_request.draft }}
           # Same count as PR preflight so transform cache shard keys line up.
-          ORCA_UNIT_FULL_SHARD_COUNT: ${{ vars.ORCA_UNIT_FULL_SHARD_COUNT || '5' }}
+          ORCA_UNIT_FULL_SHARD_COUNT: ${{ vars.ORCA_UNIT_FULL_SHARD_COUNT || '12' }}
         run: node config/scripts/ci-unit-plan.mjs
       - uses: actions/upload-artifact@v7
         continue-on-error: true

config/scripts/ci-dependency-download-cache.test.mjs

@@ -417,8 +417,8 @@ describe('CI dependency download caches', () => {
     expect(warm['runs-on']).toBe('${{ matrix.runner }}')
     expect(warm.strategy.matrix.runner).toEqual([
       'blacksmith-4vcpu-ubuntu-2404-arm',
+      pr.jobs.preflight['runs-on'],
       pr.jobs.test.with.runner,
-      'ubuntu-24.04-arm',
       'ubuntu-22.04'
     ])
     expect(workflow('ci-cache-warmup').jobs.warm.strategy.matrix.runner).toEqual([
@@ -426,7 +426,7 @@ describe('CI dependency download caches', () => {
       'ubuntu-24.04'
     ])
     expect(pr.jobs.preflight['runs-on']).toBe('blacksmith-8vcpu-ubuntu-2404')
-    expect(pr.jobs.test.with.runner).toBe('blacksmith-8vcpu-ubuntu-2404')
+    expect(pr.jobs.test.with.runner).toBe('ubuntu-24.04-arm')
     expect(units['runs-on']).toBe('${{ inputs.runner }}')
     expect(JSON.parse(pr.jobs.test.with.node_versions)).toEqual(['24'])
     expect(warmInstall.if).toBeUndefined()

config/scripts/ci-unit-plan.mjs

@@ -8,9 +8,9 @@ import { readTimingBaseline, writeAssignment } from './ci-shard-assignment.mjs'
 
 export function prepareUnitPlan(env = process.env) {
   const requestedCount = env.ORCA_UNIT_FULL_SHARD_COUNT ?? String(FULL_SHARD_COUNT)
-  // Every consumer reads the count from the plan; ten bounds runner-slot use.
-  if (!/^([2-9]|10)$/.test(requestedCount)) {
-    throw new Error('ORCA_UNIT_FULL_SHARD_COUNT must be an integer from 2 to 10')
+  // Every consumer reads the count from the plan; sixteen bounds runner-slot use.
+  if (!/^([2-9]|1[0-6])$/.test(requestedCount)) {
+    throw new Error('ORCA_UNIT_FULL_SHARD_COUNT must be an integer from 2 to 16')
   }
   const fullShardCount = Number(requestedCount)
   const files = discoverUnitFiles()

config/scripts/ci-unit-plan.test.mjs

@@ -86,7 +86,7 @@ it.each([
   }
 )
 
-it.each(['5', '6', '10', 'invalid', '11'])(
+it.each(['5', '6', '16', 'invalid', '17'])(
   'validates the requested full shard count and keeps fallback coverage: %s',
   (requestedCount) => {
     const root = mkdtempSync(join(tmpdir(), 'orca-unit-count-'))
@@ -105,10 +105,10 @@ it.each(['5', '6', '10', 'invalid', '11'])(
           ORCA_SHARD_SOURCE_SHA: 'full-count-fixture'
         }
       })
-      if (['invalid', '11'].includes(requestedCount)) {
+      if (['invalid', '17'].includes(requestedCount)) {
         expect(result.code).not.toBe(0)
         expect(result.stderr).toContain(
-          'ORCA_UNIT_FULL_SHARD_COUNT must be an integer from 2 to 10'
+          'ORCA_UNIT_FULL_SHARD_COUNT must be an integer from 2 to 16'
         )
         return
       }

config/scripts/ci-unit-selection.mjs

@@ -54,7 +54,7 @@ export function selectUnitFiles(files, changed, graphInput) {
 
 // Job slots, not cores, drive queueing: eight shards consumed 68% of daily slots. Five budget ~8.8min/shard versus 10.5min p95 queue.
 // Full PR/daily callers use five; selected drafts may use fewer.
-export const FULL_SHARD_COUNT = 5
+export const FULL_SHARD_COUNT = 12
 
 export function planUnitSelection({
   files,

config/scripts/ci-unit-selection.test.mjs

@@ -167,7 +167,7 @@ describe('conservative unit selection', () => {
     )
   })
 
-  it('uses ten complete PR shards while selected drafts retain the five-shard cap', () => {
+  it('uses ten complete PR shards while selected drafts stay within the full-count cap', () => {
     const expandedSources = new Map(sources)
     for (let index = 0; index < 5; index++) {
       expandedSources.set(`src/additional-${index}.test.ts`, `import './leaf'`)
@@ -196,7 +196,7 @@ describe('conservative unit selection', () => {
       mode: 'selected',
       event: { pull_request: { draft: true } }
     })
-    expect(selected.shards).toHaveLength(5)
+    expect(selected.shards).toHaveLength(Math.min(FULL_SHARD_COUNT, selected.executionFiles.length))
     expect(selected.executionFiles).toEqual(
       selectUnitFiles(expandedFiles, ['src/leaf.ts'], expandedGraph).files
     )

config/scripts/pr-preflight-gates.test.mjs

@@ -38,7 +38,7 @@ it('shares one setup and overlaps the compiler only after type-aware checks fini
     'node config/scripts/ci-unit-plan.mjs'
   ])
   expect(plan.env.ORCA_UNIT_SELECTION_MODE).toContain('vars.ORCA_UNIT_SELECTION_MODE')
-  expect(plan.env.ORCA_UNIT_FULL_SHARD_COUNT).toBe("${{ vars.ORCA_UNIT_FULL_SHARD_COUNT || '5' }}")
+  expect(plan.env.ORCA_UNIT_FULL_SHARD_COUNT).toBe("${{ vars.ORCA_UNIT_FULL_SHARD_COUNT || '12' }}")
   expect(steps.indexOf(plan)).toBeLessThan(compilerJoin)
   expect(steps.indexOf(compiler)).toBeGreaterThan(
     steps.findIndex((step) => step.wait === 'changed-code-quality')

config/scripts/pr-unit-support-routing.test.mjs

@@ -7,7 +7,7 @@ import {
   unitConsumers
 } from './ci-unit-dependency-graph.mjs'
 import { CROSS_VERSION_WIRE_DIR, discoverUnitFiles } from './ci-unit-files.mjs'
-import { planUnitSelection } from './ci-unit-selection.mjs'
+import { FULL_SHARD_COUNT, planUnitSelection } from './ci-unit-selection.mjs'
 
 const harness = 'src/renderer/src/runtime/web-session-tabs-sync-test-harness.ts'
 const structuredSessionHarness =
@@ -123,7 +123,7 @@ describe('unit support import ownership', () => {
     })
     expect(plan.mode).toBe('shadow')
     expect(plan.executionFiles).toEqual(files)
-    expect(plan.shards).toHaveLength(5)
+    expect(plan.shards).toHaveLength(FULL_SHARD_COUNT)
   })
 
   it.each([

docs/readme/ci-runners.md

@@ -1,10 +1,12 @@
 # CI runners
 
 Default to free GitHub-hosted standard runners: the repository is public and the
-Enterprise plan allows 500 concurrent jobs (50 macOS). Use Blacksmith only for:
+Enterprise plan allows 1,000 concurrent jobs (50 macOS). Use Blacksmith only for:
 
-- PR critical-path jobs: unit test shards, preflight and typecheck, and the layout
-  oracle gate (`e2e.yml` with `fast_runners: true`). Advisory e2e lanes run hosted.
+- PR critical-path jobs: preflight and typecheck, and the layout oracle gate
+  (`e2e.yml` with `fast_runners: true`). Advisory e2e lanes run hosted.
+  Unit tests run as 12 shards on hosted `ubuntu-24.04-arm`, which match the old
+  five 8-vCPU Blacksmith shards; `ORCA_UNIT_FULL_SHARD_COUNT` tunes the count.
 - `pr.yml` Git compatibility: its parallel real-Git lanes, two through Docker,
   exceed test timeouts on 4-vCPU hosted runners.
 - macOS packaging: adhoc, hourly, daily and `release-mac-build.yml`. Hosted