Checks

Every intervalSeconds, baba reads each enabled check once, one after the other, and logs a status line:

CPU: 12% | Load: 0.52 | Memory: 41% | Disk: / 63%, /data 80% | GPU: N/A

A reading is a breach when it's strictly above its threshold. What happens next (consecutive breaches, reminders, recovery) is in Incidents. A check that can't read its metric logs an error and is skipped for that cycle; the others still run.

CPU

Percentage of CPU time spent busy (everything but idle and I/O wait) over a one-second sample taken during the cycle, rounded to a whole number. On macOS, the second of two top samples one second apart.

Config: checks.cpu; alert: ⚠️ **CPU LOAD**: Usage is at **95%**.

Load

The 1-minute load average: how many processes were running or waiting for CPU. Compare it to your core count; a threshold equal to the number of cores alerts when the machine is saturated.

Config: checks.load; alert: 🚨 **LOAD CRITICAL**: Load average is at **9.12**.

Memory

Used memory is total minus available: page cache and buffers the kernel can reclaim count as free, like htop and free's "available" column. On Linux, MemAvailable (or free + buffers + cached on old kernels); on macOS, free + inactive + speculative pages.

Config: checks.memory; alert: ⚠️ **MEMORY USAGE**: Usage is at **93% (14.88 GB/16.00 GB)**.

Disk

For every mounted filesystem whose mount point or device is in checks.disk.volumes, used space is size minus the space available to regular users. Reserved blocks (ext4 keeps 5% for root) therefore count as used, like df; on macOS, APFS volumes sharing a container all report the container's free space.

A filesystem mounted several times (bind mounts) is checked once. If nothing in the list is mounted, the status says Disk: no volumes found. Each volume is its own incident, keyed by device, and opens on the first breach: there is no consecutiveBreaches for disks, since disk usage doesn't spike and fall back.

Config: checks.disk; alert: ⚠️ **DISK USAGE** (/dev/sda1): Usage is at **92% (430.12 GB/468.00 GB)**.

In Docker, see Disks in a container.

Temperature

• CPU (Linux): the hottest temp*_input of the hwmon sensors named coretemp, k10temp, zenpower, cpu_thermal, cpu-thermal or soc_thermal. Without any, the hottest thermal zone whose type mentions cpu, x86_pkg or soc. Readings at or below 0 are ignored.
• GPU: each NVIDIA GPU's temperature from nvidia-smi, as its own incident.

macOS has no CPU source without private frameworks. With no reading at all, the check adds nothing to the status line.

Config: checks.temperature; alert: 🌡️ **CPU TEMP**: Temperature is at **91°C**.

GPU VRAM

Memory used over memory total for each NVIDIA GPU, from nvidia-smi --query-gpu=name,memory.used,memory.total,temperature.gpu. Each GPU is its own incident, keyed by name. Other vendors aren't supported; without nvidia-smi, the status says GPU: N/A.

Config: checks.gpu; alert: ⚠️ **GPU VRAM** (NVIDIA GeForce RTX 3090): Usage is at **95%**.