banshee/linux-collection/linux/ssh/filesystems

Filesystems of a Linux, by SSH

Returns, for each mounted filesystem, the used space, the free space, the size, the used and free inodes, whether it is mounted read-only, and whether the account could read it. Memory and pseudo filesystems are left out. Seven watches are offered, each one set on the probe.

Versions

Generic: /proc/mounts is written by the kernel the same way on Alpine, Debian and Fedora; -P, -k and -i are options of df in BusyBox as in GNU coreutils. A df that does not know -i (a BusyBox built without it) stops the collector misconfigured.

How it works

The collector runs three commands:

cat /proc/mounts, for the type and the mode (rw or ro) of each mount;
df -P -k, for the space, in KiB;
df -P -i, for the inodes.

A mount point shown twice (stacked mounts) is judged once. A filesystem of size zero (proc, sysfs, cgroup…) is left out, as are the types tmpfs, devtmpfs, ramfs, overlay, squashfs, iso9660, udf, efivarfs, fuse.lxcfs, fuse.gvfsd-fuse and fuse.portal (the last two, views a desktop session mounts for its user alone).
A mount the account may not read (the network or FUSE mount of another user, a mount point without the right to enter it) is named by df on its error output (df: /mnt/x: Permission denied). It is returned with its mode and access at unreadable, and nothing else: no space, no inodes. A readable filesystem has access at readable.
The used share is the one df shows: used over used plus available, root's reserve excluded. The free space is the space available to an ordinary account.
An ext4 that meets an error remounts itself read-only: the mount mode is read for that.
A filesystem that does not count inodes (btrfs, vfat) has no inode values: a filesystem can be full with free space left, when its inodes run out.
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

Name	Label	Unit	Per instance
`used-percent`	Used space	%	yes, per filesystem (its mount point)
`used`	Used space	GiB	yes, per filesystem
`free`	Free space	GiB	yes, per filesystem
`size`	Size	GiB	yes, per filesystem
`inodes-used`	Used inodes	%	yes, per filesystem
`inodes-free`	Free inodes	—	yes, per filesystem
`mode`	Mount mode: `rw` or `ro`	—	yes, per filesystem
`access`	Access: `readable`, or `unreadable` when the account may not read it	—	yes, per filesystem

Watches

Name	Title	Kind	Defaults	Checked by default
`used`	Used space	above a threshold, on `used-percent`	warning above 80 %, critical above 90 %	yes
`full-soon`	Full within a week	above a threshold, on `used-percent`	warning above 99 %, ahead 7 d	yes
`free`	Little free space	below a threshold, on `free`	critical below 5 GiB; no warning	no
`inodes`	Inodes running out	above a threshold, on `inodes-used`	warning above 80 %, critical above 90 %	yes
`read-only`	Mounted read-only	state, on `mode`	critical on `ro`	yes
`unreadable`	Filesystem unreadable	state, on `access`	warning on `unreadable`	yes
`missing`	Filesystem missing	expected instances, on `used-percent`	no filesystem named: it raises nothing until names are set on the probe	no

Limits

A stale network mount can make df hang: the run then ends on the SSH timeout, unreachable.
In a mount point, a run of blanks is read as a single one.
A filesystem mounted read-only on purpose raises read-only: uncheck it, or leave it out of the selection. Likewise a mount the account is not meant to read raises unreadable: leave it out of the selection.
A mount is told unreadable from the line df writes for it (df: <point>: <reason>, the point between quotes or not): a df that writes it otherwise leaves the mount out.