banshee/linux-collection/linux/ssh/temperatures

Temperatures of a Linux, by SSH

Returns the temperature of each sensor the kernel exposes, and raises when a sensor crosses the thresholds it announces itself. A virtual machine has no sensor: the collector says so.

Versions

Generic: /sys/class/hwmon and /sys/class/thermal are written by the kernel the same way on Alpine, Debian and Fedora; -s and -H are options of BusyBox as of GNU grep.

How it works

The collector reads with grep -s -H the files of /sys/class/hwmon: for each chip its name, and for each sensor its reading (temp*_input), its label, its high (max) and critical (crit) thresholds, in thousandths of a degree. Failing any sensor there, it reads the thermal zones of /sys/class/thermal (type and temperature), the only ones on some ARM boards.

A sensor is named after its chip and its label (coretemp Package id 0), or its number when the driver gives no label (nvme temp1); two sensors of the same name each keep their own, the second carrying its chip.
Thermal zones are not read beside hwmon: they often repeat its sensors.
Nothing is written on the target.

Prerequisites

an SSH account for monitoring, without write rights;
the fingerprint of the server's host key, set in protocol:ssh:fingerprint (not needed when the Security settings trust the first SSH fingerprint presented);
the drivers of the sensors loaded (coretemp, k10temp, nct6775…): lm-sensors is not needed.

Outputs

Name	Label	Unit	Per instance
`temperature`	Temperature	°C	yes, per sensor

Watches

Name	Title	Kind	Defaults	Checked by default
`hot`	High temperature	above a threshold, on `temperature`	warning above 80 °C, critical above 95 °C, for 3 runs	yes
`sensor-limit`	Sensor threshold crossed	custom	warning at the sensor's `max`, critical at its `crit`	yes

Limits

A machine without any sensor (a virtual machine, a container) stops the collector misconfigured.
What is hot depends on the part: a laptop CPU at 90 °C is normal, a disk at 60 °C is not. The defaults of hot suit a server CPU; sensor-limit uses the part's own thresholds.
A driver may announce no threshold, or a wrong one: sensor-limit then judges nothing, or too late.