mirror of https://github.com/openobserve/openobserve.git synced 2026-11-05 08:09:22 +00:00

Files

T

Vamsi Karthik RandShrinath Rao 6930a9b724 test(mobile): RUM SDK end-to-end suite across 4 platforms + self-contained CI (#13628 )

## What this PR adds

end-to-end test suite for the **Mobile RUM SDK** under
`tests/mobile-testing/`, plus a **self-contained CI workflow** that runs
it on real emulators/simulators against an OpenObserve **built from this
PR's source**. Test / fixture / CI only — no product-code/existing test
changes

## How each test works (3 layers)

Every scenario is a real end-to-end journey, so a failure points at the
actual broken layer:

1. **Drive** — a Maestro flow taps through the app like a user
(`maestro/<track>/*.yaml`).
2. **API** — query the `_rumdata` stream and assert the telemetry is
correct (`utils/ooClient.js`, poll-and-retry for async ingestion).
3. **UI** — open the RUM dashboard and assert the sessions

## Coverage (42 tests across 4 platforms)

The **native platforms (android-native, ios-native) are now at parity
with React Native** — each new
native test (masking, network, bg/fg, user-identity, no-phone-home) was
built + validated on a real
emulator/simulator before landing.

| Capability | Layer | RN-Android | Android-native | RN-iOS | iOS-native
|
|---|---|:-:|:-:|:-:|:-:|
| Core telemetry (views/actions/handled-error/crash → `_rumdata`) | API
| ✅ | ✅ | ✅ | ✅ |
| Crash session viewable in dashboard | UI | ✅ | ✅ | ✅ | ✅ |
| Attributes: env / service / version tagging | API | ✅ | ✅ | ✅ | ✅ |
| User identity (`setUser`) | API | ✅ | ✅ | ✅ | ✅ |
| Network: fetch → resource + 4xx/5xx | API | ✅ | ✅ | ✅ | ✅ |
| Session-replay PII masking | UI | ✅ | ✅ | ✅ | ✅ |
| Background / foreground view transitions | API | ✅ | ✅ | — | ✅ |
| No-phone-home (no egress to Datadog hosts) | security | ✅ | ✅ | — | ✅
|

- Shared per-capability factories keep the specs DRY
(`utils/rumChecks.js`, `utils/coreRumSpec.js`).
- Negatives are covered: 4xx capture, PII-must-not-leak (masking), no
phone-home, handled-error ≠ crash.
- `o2-enterprise#2289` dashboard bugs are kept as inline skips that flip
green once fixed.
- Full capability × platform detail + on-device status:
`docs/COVERAGE-MATRIX.md`.

## Fixture apps + build-time config

- Four vendored apps under `apps/`: `o2-rum-tester` (React Native —
drives both RN platforms), `o2-native-android` (Kotlin), `o2-native-ios`
(Swift / xcodegen).
- Each reads its RUM target (endpoint / org / token) and SDK version
from a **build-time-overridable** config. The committed default is the
current pin, so local dev is unchanged; CI regenerates it to point at
the runner's OpenObserve and can pin the latest SDK on release.

## CI — `.github/workflows/mobile-sdk-e2e.yml` (self-contained, no
secrets)

- **Build once per OS from source.** `build_oo_linux` / `build_oo_macos`
compile OpenObserve (binary + embedded web UI) from this PR's code and
cache it. Device jobs download and run it **natively**, and the app
posts real RUM to it (Android reaches the host at `10.0.2.2`, the iOS
simulator via `localhost`). No Docker, no shared cluster, no secrets.
- **Device jobs run in parallel.** `android` (Linux + KVM emulator →
rn-android + android-native) and `ios` (macOS simulator, pinned Xcode
16.4 → rn-ios + ios-native).
- **`merge_reports`** combines both platforms into one browsable HTML
report — even when a device job fails.
- **`decide` picks which platform runs, to keep scarce macOS runners in
check:**
- PR touching Android or shared mobile-RUM code → **Android auto-runs**;
iOS is gated behind the `mobile-e2e` label.
  - `mobile-e2e` label → both platforms (full pre-merge gate).
- `repository_dispatch(mobile-sdk-release)` from an SDK repo → runs that
SDK's platform (companion PRs wire this up:
openobserve-react-native-rum#3, openobserve-sdk-android#1,
openobserve-sdk-ios#1).
  - `workflow_dispatch` → manual, choose platforms.

*Separate workflow :* it needs macOS **and** Android emulators, builds 4
phone apps, and is dispatch-driven — matching the repo's existing
per-suite split

## Status

Green end-to-end on GitHub runners (Android + iOS device jobs + merge) :
https://github.com/openobserve/openobserve/actions/runs/31733717758/job/94565520891

---------

Co-authored-by: Shrinath Rao <shnath@openobserve.ai>

2026-08-20 08:13:35 +00:00

6.9 KiB

Raw Permalink Blame History

Mobile RUM SDK — Coverage Matrix & Gap Analysis

Coverage of the tests/mobile-testing suite across the four mobile tracks. 42 tests, 5 Playwright projects (rn-android, android-native, rn-ios, ios-native, sdk-drift). The native tracks are now at RN parity — masking, network, background/foreground, user-identity and no-phone-home ship for both android-native and ios-native (previously listed here as deferred). The PR-description matrix is the authoritative capability×platform view; anything below marked "deferred / N/A" for those native capabilities is superseded by the shipped specs.

How the suite is structured

DRIVE → API → UI on every scenario: Maestro taps the app (maestro/), the OpenObserve Search API proves the data landed (utils/ooClient.js), the RUM dashboard proves it renders (pages/rumDashboardPage.js).
Per-capability factories (utils/rumChecks.js) keep the four platforms at parity without copy-paste: attributesSuite, userIdentitySuite, networkSuite (positive + 404 negative), maskingSuite, bgFgSuite, noPhoneHomeAndroidSuite. Each platform's spec is a thin call.
coreRumSpec.js bundles the end-to-end crash journey (views + actions + handled error + native crash in _rumdata, then the crashed session rendering in the dashboard) for the three non-RN-Android platforms.

The RUM capability × platform matrix

Legend: ✓ covered · N/A app can't do it without a code change · ⏭ deferred (see follow-ups) · ⚠ blocked by dashboard bug o2-enterprise#2289.

Capability (layer)	RN-Android	Android-native	RN-iOS	iOS-native
Session created (API)	✓	✓	✓	✓
Named view tracking (API)	✓	✓ bundle	✓ bundle	✓ bundle
Actions — tap/custom (API)	✓ bundle	✓ bundle	✓ bundle	✓ bundle
Resource: fetch success (API)	✓	N/A no net	✓	N/A no net
Resource error: 4xx status (API, negative)	✓	N/A	✓	N/A
Handled error, is_crash=false (API)	✓	✓ bundle	✓ bundle	✓ bundle
Native crash, is_crash=true (API)	✓	✓	✓	✓
User identity — usr_* fields (API)	✓	✓	✓	N/A app doesn't set
Attributes: env/service/version (API)	✓	✓	✓	✓
Session replay privacy masking (UI, negative)	✓	N/A no form	✓	N/A no form
Background/foreground continuity (API)	✓	⏭	⏭	⏭
No-phone-home: no Datadog hosts (security, negative)	✓	✓	N/A IPA	N/A IPA
Crashed session viewable (UI)	✓	✓	✓	✓
Error Tracking tab shows error (UI)	⚠ skip	⚠ skip	⚠ skip	⚠ skip

On-device validation status (Phase 5)

Run against the emulator + booted iOS simulator on the migrated dev cluster:

Validated green on-device: RN-Android attributes / user-identity / network incl. the 404 negative / no-phone-home / background-foreground; Android-native attributes / user-identity / no-phone-home; RN-iOS attributes (confirms iOS-through-the-factory + simulator device targeting).
Generated + statically validated, pending a device run: RN-iOS network (+404) / user-identity / masking; iOS-native attributes; plus the pre-existing per-platform core bundles and the RN-Android crash / masking / views / handled-error specs. (The suite parses and lists cleanly; they reuse the exact factory paths already proven above, so they need a booted sim to execute, not new logic.)

Key finding: the SDK does capture resource_status_code for failing requests — the 404 negative asserts >= 400 and passes on RN-Android.

Deferred / follow-ups (not shipped as tests)

Background/foreground on native (Android-native, iOS-native) and RN-iOS. The RN-Android adb cycle (HOME → am start) keeps one session and passes. The same mechanism produced two sessions on Android-native (activity-based views), and the iOS Home-key backgrounding was unreliable on the simulator. Needs investigation: is native session continuity across background a real SDK difference, or a drive-mechanism artifact? Held out rather than shipped red/flaky.
Native network / masking tests — the Android-native and iOS-native apps make no network calls and have no PII form; needs a Fetch button + a form added to those fixture apps.
iOS-native user identity — that app never calls setUserInfo (one-line app change, or accept N/A).
iOS no-phone-home — scanning an .ipa for Datadog hosts is materially harder than the Android APK scan; separate follow-up.
Error Tracking tab (o2-enterprise#2289) — kept as a test.fixme skip on every platform.

Negative-scenario coverage

Shipped negatives: 404/4xx resource status captured (RN, validated), PII must not leak into the replay (masking), no Datadog hosts in the app (no-phone-home), handled error must be is_crash=false (all platforms). Offline/airplane-mode behavior remains a possible later addition.

Product findings (log to o2-enterprise#2289)

The RUM install docs in the UI pin a STALE SDK version. The Data Sources → Real User Monitoring page has all four platform tabs (Browser · React Native · Android · iOS, all BETA) — so the docs exist. But the versions are hardcoded and have drifted behind the shipped SDKs:

Platform	Version shown in the UI	Version we build/test	Drift
React Native	`0.1.0-alpha.4` (`RUM_RN_SDK_VERSION`, OSS `rumReactNative.ts`)	`alpha.6`	2 behind
Android	`0.1.0-alpha4` (enterprise card)	`alpha5`	1 behind
iOS	`from: "0.1.0-alpha.4"` (enterprise card)	`alpha.5`	1 behind
A customer copy-pasting from this page installs an older SDK than what's published. Root cause: the
version is a hardcoded constant with no freshness check.

The mobile RUM SDK is NOT enterprise-gated — corrected. RUM ingestion (/rum/v1/... in the base OSS router, no enterprise import), the RUM dashboard (web/src/views/RUM/*), and the SDK packages (public npm/Maven/GitHub) are all open source. The ONLY edition difference is which install-doc cards are surfaced: OSS FrontendRumConfig.vue ships PLATFORMS = [browser, reactNative] (code comment: "Adding iOS / Android / Flutter later"), while the enterprise build additionally shows the Android + iOS cards. So a docs-presence/drift test for Android/iOS runs @enterprise, but the SDK functionality tests are all OSS — which is why the self-contained (plain-OSS) CI works.
How to verify (proposed): (1) a drift test comparing the UI-pinned version against the latest published (npm / Maven / GitHub releases) — currently RED; (2) executable docs — this E2E suite already proves "these coordinates + this init code produce working RUM"; with the new version injection it can run against the documented version and the latest to prove both work.

6.9 KiB Raw Permalink Blame History Unescape Escape