Admiral v00047: Low-Level Watchdogs, Offline Telemetry Buffers, and Power Metrics

Admiral v00047 focuses on hard hardware-level survivability, deep offline tolerance for remote spatial sensors, and enhanced power profiling for mobile edge nodes.

Here is a breakdown of what has landed.

1. Physical Hardware Watchdogs & Boot Safety

When edge nodes sit in remote or physically inaccessible locations, operating system deadlocks must be recovered by the silicon itself. We have introduced deep, low-level safety loops:

• Hardware Watchdog Petting: The host supervisor now natively pets /dev/watchdog. If the core agent runtime deadlocks or fails to cycle, the physical hardware will hard-reboot the machine.
• Aggressive Boot Guard Timers: We have tightened connectivity check windows to get devices back online faster. On a fresh first-time boot, the system now reboots after 1 hour of no connectivity (down from 6 hours). For running nodes, the automated non-communication reboot threshold has been pulled in from 7 days to 3 days.
• Emergency Update Recovery: Introduced an out-of-band rescue and emergency update flow inside the Provisioning API to recover partially bricked or un-paired units.
• Persistent Filesystem Recovery: Resolved an onboarding issue by attempting to repair and resume the existing local filesystem on a reboot, rather than wiping and writing a fresh layout.

2. AirDetect Offline Buffering & BLE Refinements

Our spatial analytics engine can now survive long-term WAN dropouts without losing critical footprint data:

• Metrics Caching & Buffering: AirDetect telemetry metrics are now cached and buffered locally during offline periods, flushing smoothly once WAN paths recover.
• Decoupled BLE Provisioning: Reworked our Bluetooth provisioning flows to run reliably even if the primary AirDetect analytics engine is disabled.
• BLE Spec Fixes: Handled structural edge cases inside the local BLE spec parser to prevent malformed advertising frames from interrupting pairing loops.

3. Hardened Networking & Control Plane Optimization

We continue to reduce management-overhead noise on expensive satellite and cellular uplinks:

• No More Latency Monitor Overhead: Removed the dedicated latency monitoring service. Ping and link quality diagnostics are now multiplexed into existing, low-overhead streams.
• Targeted SD-WAN Handshakes: The SD-WAN path-selection logic now attempts to ping our core API server first, verifying access to our most critical control interface before routing ambient traffic.
• Traffic Grace Periods: Added startup grace periods and increased sender timeouts across metrics and log pipelines to prevent boot-up cascades from polluting control sockets.
• State Preservation: Integrated a dedicated network connection store to prevent interface transitions from dropping active connection maps.

4. Workload Isolation, Screens, & Power Profiles

• Low-Power Battery Metrics: The system now automatically detects hardware battery interfaces, exposing charge states, voltage, and drain telemetry profiles in native system metrics.
• FFmpeg-Backed Screenshots: Screen visual diagnostics now utilize FFmpeg captures, adding support for RK3326-based displays.
• Workload Startup Fixes:
  • Workloads now start correctly on hosts lacking AppArmor.
  • Errors are no longer swallowed during initial workload execution.
  • Workload target configurations are only applied when a functional difference or update is detected.
  • Fixed image pulling failures on offline boots.

5. TUI Adjustments

• Compact Layouts: Added a smaller, high-density Terminal User Interface (TUI) layout suited for small-screen visual terminals.
• Battery Support: Real-time physical battery levels are now rendered directly on the local terminal status head.