Internals: the health monitor

kernel/core/core_health.{c,h} - 8 declarations; the numbered 5:

health_get_cpu(out)
health_get_mem(out)
health_get_tasks(out)
health_get_io(out)
health_export(path)

What the metrics see

• CPU: per-core busy accounting over a 1 s window - three spinner tasks measurably pin the core (100% observed live; >= 90% sustained across 10 consecutive samples is the alarm threshold).
• Memory: PMM/heap state.
• Tasks: process table census.
• IO: block/net counters.

Alarms

Threshold crossings write to the kernel log: CPU-LOCKED when a core stays pinned (the E2E S11 runs 3 spinners and asserts the alarm while the kernel responds), memory pressure, IO stall.

Export

health export writes valid JSON to /var/log/health.json (386-389 bytes of the standard metric set in the E2E) - parseable by any monitoring tool, including from the /etc volume after a reboot.

Commands

health (summary), health watch (1 s live), health export.