ci(e2e-v2): New test servers (e2e-test-stack) for mobile CI

[saturninoabril]

Summary

Adds an independent E2E v2 pipeline that does not modify the existing Matterwick workflows (e2e-detox-pr.yml, e2e-maestro-pr.yml, and friends).

v2 leases its own test servers through e2e-test-stack-actions. Each platform has its own create/destroy lifetime (separate batch, hostname prefix, and commit statuses). Stacks are leased only after the matching app artifact succeeds. A failed iOS simulator build skips iOS / iPad / Maestro iOS create+tests; Android still runs if its build passed.

Phase 0 scope: PR only. Templates take run_type / context_suffix / per-platform parallelism so main, release, and CMT can reuse them later.

Detox gates:

• e2e-test/detox-ios: full phone suite (empty include tags, exclude @ipad_only), 30 workers, stack 30+2
• e2e-test/detox-android: exclude @ipad_only,@ios_only, 20 workers, stack 20+2
• e2e-test/detox-ipad: include @ipad_only, 1 worker, stack 1 (SITE_1 only)

Maestro gates:

• e2e-test/maestro-ios: 1 worker, stack 1 (SITE_1 only), reuses the iOS simulator artifact
• e2e-test/maestro-android: 1 worker, stack 1 (SITE_1 only), uses the Maestro Android APK (android-maestro-build-files-*)

workflow_dispatch parallelism overrides iOS and Android Detox only; iPad and Maestro stay at 1.

flowchart TD
  PR[pull_request TEMPORARY] --> Wait[wait-for-ci]
  UI[workflow_dispatch Actions UI] --> Wait
  Wait --> DetoxCheck["e2e-test/detox-check"]
  Wait --> MaestroCheck["e2e-test/maestro-check"]
  Wait --> BuildIOS["e2e-test/ios-build or ios-build-cache"]
  Wait --> BuildAndroid["e2e-test/android-build or android-build-cache"]
  Wait --> BuildMaestroAndroid["e2e-test/android-maestro-build or android-maestro-build-cache"]
  BuildIOS --> CreateIOS["create ios stack"]
  BuildIOS --> CreateIpad["create ipad stack"]
  BuildIOS --> CreateMaestroIOS["create maestro-ios stack"]
  BuildAndroid --> CreateAndroid["create android stack"]
  BuildMaestroAndroid --> CreateMaestroAndroid["create maestro-android stack"]
  DetoxCheck --> DetoxIOS[detox-ios]
  DetoxCheck --> DetoxAndroid[detox-android]
  DetoxCheck --> DetoxIpad[detox-ipad]
  MaestroCheck --> MaestroIOS[maestro-ios]
  MaestroCheck --> MaestroAndroid[maestro-android]
  CreateIOS --> DetoxIOS
  CreateAndroid --> DetoxAndroid
  CreateIpad --> DetoxIpad
  CreateMaestroIOS --> MaestroIOS
  CreateMaestroAndroid --> MaestroAndroid
  DetoxIOS --> DestroyIOS["destroy ios stack"]
  DetoxAndroid --> DestroyAndroid["destroy android stack"]
  DetoxIpad --> DestroyIpad["destroy ipad stack"]
  MaestroIOS --> DestroyMaestroIOS["destroy maestro-ios stack"]
  MaestroAndroid --> DestroyMaestroAndroid["destroy maestro-android stack"]

Loading

Ticket Link

E2E test automation v2 (no Matterwick).

Checklist

• Added or updated unit tests (required for all new features)
• Has UI changes
• Includes text changes and localization file updates
• Have tested against the 5 core themes to ensure consistency between them.
• Have run E2E tests by adding label E2E/Run (or E2E/Run-iOS / E2E/Run-Android for platform-specific runs).

Device Information

This PR was tested on: GitHub Actions workflows (YAML + Jest). Live Detox / Maestro runs on the E2E v2 workflow after CI is green.

Screenshots

N/A — CI workflow only.

Release Note

NONE

Follow-up on CI

Root Jest picks up .github/actions/**/*.test.js. The v2 helper tests are now Jest suites (same pattern as compute-e2e-build-fingerprint/paths.test.js) so ci / test stays green.

Follow-up on detox-ios subset (run 31996338717, 110 passed / 4 failed)

account.e2e.ts MM-T5114_3 failed because email notifications were disabled, so the option was missing: Email notifications disabled by System Administrator.

• Fix: set MM_EMAILSETTINGS_SENDEMAILNOTIFICATIONS=true on the v2 stack and patch the same flag during provision.

global_threads.e2e.ts MM-T4805_4 expected hardcoded admin, but the test-stack starter is sysadmin: Global threads starter display name is sysadmin.

• Fix: assert adminUsername from test config (workers already set ADMIN_USERNAME from stack credentials). MM-T4805_5 was a cascade from that failure.

Phase 6

The @ios_pr / 8-worker subset is green, so PR defaults moved to the full phone suite. Worker counts are now iOS 30 / Android 20 / iPad 1. workflow_dispatch can still override iOS and Android Detox parallelism (iPad and Maestro stay at 1).

Instances from create artifact

Create no longer passes the instances JSON through reusable-workflow outputs. Masked admin passwords caused GitHub to redact that output, so workers saw INSTANCES as empty. Workers now download e2e-create-<batch_id> and bind instances[shard-1] as SITE_1. Phone stacks also bind the last two entries as SITE_2 / SITE_3. iPad and Maestro bind SITE_1 only.

instance_key

Per-platform hostname prefixes (toolkit limit 24 characters): mobile-pr-<n>-ios, mobile-pr-<n>-and, mobile-pr-<n>-ipad, mobile-pr-<n>-mios, mobile-pr-<n>-mand (android is and so mobile-release-cut-and fits). Same pattern for main / release / release-cut.

Follow-up: create workflow parse error

GitHub rejected e2e-stack-create-v2.yml at parse time (Unexpected symbol: '+') because expressions cannot add fromJSON(...) values. Instance count is now $((WORKERS + EXTRAS)) in bash.

Follow-up: temporary pull_request trigger

workflow_run on ci never started E2E on this PR because that trigger always uses the default-branch workflow file, and e2e-pr-v2.yml is not on main. TEMPORARY: listen on pull_request so remaining phases can run (Actions job rows will appear). workflow_dispatch stays so the GitHub Actions UI can run a chosen ref. Before review: switch back to workflow_run on ci completed and keep workflow_dispatch.

Follow-up: Node 24 action pins

ci.yml, prepare-node-deps, and test-coverage now pin checkout / setup-node / cache / artifacts to current Node 24 SHAs. Coverage PR comments use actions/github-script instead of thollander/actions-comment-pull-request (still Node 20).

Follow-up: S3 build cache on ubuntu-24.04

Fingerprint + S3 restore for iOS/Android artifacts runs on ubuntu-24.04. macos-26 / ubuntu-latest-8-cores start only on a cache miss.

Follow-up: lease stacks after builds

Create jobs now needs the matching build. iOS simulator success gates Detox iOS, Detox iPad, and Maestro iOS. Android Detox APK gates Detox Android. Maestro Android APK gates Maestro Android.

Follow-up: Maestro first run (run 32140040219)

Android: 5/15 passed (account + clock_display green). iOS: 0 passed; worker cancelled at 60m with 1 spec missed.

Cause: exclude_tags.json no longer excluded @android_only, @ios_only, or @multi_device, so Test System IO leased Calls on iOS and six two-device flows on one worker. iOS login also tapped the LDAP placeholder (Email, Username or AD/LDAP Username); the v2 stack does not enable LDAP, so the field is Email or Username. Android uses the username testID, which is why those account flows passed.

file_type_preview (MM-T3244) ran without seed_file_preview.ts and looked for id: undefined-file-container. Android call_ui_permission hit the 9m Maestro timeout (emulator WebRTC); the next Calls/bookmark flows failed in 0s, then Maestro recovered.

Fix: restore platform + @multi_device + MM-T3244 excludes, and match iOS login with Email.*Username.*.

Existing v1 Matterwick E2E is unchanged.

  

[github-actions]

Coverage Comparison Report

Generated on August 21, 2026 at 06:25:37 UTC

+-----------------+------------+------------+-----------+
| Metric          | Main       | This PR    | Diff      |
+-----------------+------------+------------+-----------+
| Lines           |     88.39% |     88.39% |     0.00% |
| Statements      |     88.25% |     88.25% |     0.00% |
| Branches        |     77.33% |     77.33% |     0.00% |
| Functions       |     87.73% |     87.70% |    -0.03% |
+-----------------+------------+------------+-----------+
| Total           |     85.42% |     85.41% |    -0.01% |
+-----------------+------------+------------+-----------+