banshee/linux-collection/linux/ssh/load

Load average of a Linux, by SSH

Returns the load average over 1, 5 and 15 minutes, the same divided by the number of cores, and the number of cores. A load of 1 per core keeps the whole machine busy; beyond it, tasks wait for a core.

How it works

The collector runs cat /proc/loadavg; grep -c '^cpu[0-9]' /proc/stat: the three averages, then the number of online cores (one cpuN line each in /proc/stat).

Portable: both files are written by the kernel the same way on Alpine, Debian and Fedora; grep -c exists in BusyBox and GNU grep. /proc/cpuinfo is not read, its layout changes with the architecture. Nothing is written on the target.
An answer that is not two lines, or a load or core count that is not a number, stops the collector misconfigured.

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).

Outputs

Output	Unit	Meaning
`load-1`, `load-5`, `load-15`	—	load average over 1, 5 and 15 minutes
`load-1-per-core`, `load-5-per-core`, `load-15-per-core`	—	the same divided by the number of cores
`cores`	—	number of online cores

Watches

Watch	Kind	Defaults	Checked
`load-high` Machine overloaded	above, on `load-5-per-core`	warning 1.5, critical 3	yes
`load-unusual` Unusual load	habit, on `load-5-per-core`	sensitivity normal	no

Limits

Linux counts in the load the tasks waiting for a disk (state D): a high load with a low CPU usage points to storage, see linux/ssh/cpu (iowait) and linux/ssh/diskio.
In a container, the load is the host's.