banshee/linux-collection/linux/ssh/diskio

Disk activity of a Linux, by SSH

Returns, for each disk, the reads and writes per second, the read and write throughput, the mean wait of a read and of a write, and the share of time the disk was busy.

How it works

The collector runs ls /sys/block (the whole disks) and cat /proc/diskstats (the counters of each disk and partition since boot).

Only the disks of /sys/block are measured: it never lists partitions, whatever their name (sda1, nvme0n1p1, mmcblk0p1).
Every value comes from counters over the interval between two runs: the first run is skipped, and a disk whose counters went backwards is left out of that run.
The mean wait is the time spent on reads (writes) divided by their number; a disk without reads (writes) in the interval has no wait value.
Portable: both are written by the kernel the same way on Alpine, Debian and Fedora; a sector in /proc/diskstats is always 512 bytes. 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).

Outputs

Output	Unit	Per	Meaning
`read-iops`, `write-iops`	op/s	disk	reads and writes completed
`read-rate`, `write-rate`	MiB/s	disk	throughput
`read-wait`, `write-wait`	ms	disk	mean wait of an operation, queue included
`busy`	%	disk	share of time with at least one operation in progress

By default, loop*, ram*, zram*, sr*, fd* and nbd* are left out; the selection is changed on the probe.

Watches

Watch	Kind	Defaults	Checked
`read-latency` Slow reads	above, on `read-wait`	warning 50, critical 200, for 5 min	yes
`write-latency` Slow writes	above, on `write-wait`	warning 50, critical 200, for 5 min	yes
`busy` Disk saturated	above, on `busy`	warning 90, for 10 min	no

Limits

busy only means something for a disk that serves one operation at a time: an SSD or NVMe at 100 % may still serve many more. That is why it is not checked.
LVM and RAID volumes (dm-*, md*) are measured beside the disks under them: their activity is counted twice, once per layer.