fix(backend): detect RK3588 via device-tree compatible marker

[andrescera]

What

isRealDevice() (apps/backend/src/modules/system/device-detection.ts) now
also reads /proc/device-tree/compatible (case-insensitively matched for the
rk3588 marker) in addition to the existing /proc/device-tree/model check.
Either match classifies the board as a real device. A new injectable
readDeviceTreeCompatible probe is added to DeviceDetectionDeps /
defaultDeviceDetectionDeps, mirroring readDeviceTreeModel exactly.

Why

On a real, physical Radxa ROCK 5B+ (confirmed live over SSH), isRealDevice()
returned false and the device was treated as emulated. Root cause: the
only RK3588 check read /proc/device-tree/model, which on this board reads:

Radxa ROCK 5B+

No RK3588 substring anywhere, so the check failed; the x86-DMI fallback path
didn't match either, so detection fell through to false. That directly caused:

• Settings → Sources showing "Network ingest isn't available on this device in
  emulated mode" on a real device.
• The RTMP/SRT gateway probes (which are isRealDevice()-gated in
  network-ingest.ts) reporting "Service not running" even though the actual
  ceralive-rtmp-gateway.service unit was active (running).

/proc/device-tree/compatible is the reliable, standard board-identification
marker. On the same board it reads (NUL-separated on disk):

radxa,rock-5b-plus rockchip,rk3588

which reliably contains rk3588. The model check is kept as a
belt-and-suspenders fallback for boards whose model string does name the SoC
(e.g. Orange Pi 5+). The marker match was made case-insensitive so the lowercase
rockchip,rk3588 (compatible) and uppercase RK3588 (model) both hit one code
path — this is strictly more permissive, and the fail-closed cases (RK3399,
RK3568, generic Rockchip) still fail since none contain rk3588 in any case.

How to verify

cd apps/backend
bun test src/tests/device-detection.test.ts src/tests/device-detection-override.test.ts   # 30 pass
bun tsc --noEmit                                                                            # clean

A new regression test mirrors the exact field bug: model "Radxa ROCK 5B+ \n"
(no RK3588) with compatible "radxa,rock-5b-plus\u0000rockchip,rk3588\u0000"
resolves true.

Risks

Low, and additive:

• The override / isDevelopment / x86-mini-PC DMI branches are untouched.
• Fail-safe semantics are preserved — the new probe is wrapped in the same
  readOptional swallow-on-reject, so any read failure still degrades to ""
  and never propagates or crashes boot.
• Case-insensitive matching only ever adds matches; every existing fail-closed
  identity still fails closed (covered by the existing test.each cases).

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro

Run ID: 32458453-4fe7-4c60-a491-121ab5bb456b

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/rk3588-realdevice-marker

---

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.}