DiskIO on macOS

[DPattee]

Any chance of getting diskio available for Mac / darwin servers?

System Diskio Metricset
The System diskio metricset provides disk IO metrics collected from the operating system. One event is created for each disk mounted on the system.

This metricset is available on:

Linux
Windows
FreeBSD (amd64)

(I tried a forum search, didn't find anything relevant, if there is already a topic just fwd me to it)

[andrewkroh]

It has been implemented for macOS in 6.x. The change is included in 6.0.0-alpha2 release. See the change log. Please try the feature in the alpha release.

[DPattee]

I've run the alpha for a little while now and it seems relatively stable for my light load and slow polling times.

It could be interesting if the partitions were able to send data (disk0 as normal, but also disk0s1, s2, etc) but that info might not even be available in an easily consumable way haven't checked.

I'm not convinced that the Files count is right, and in some cases it is definitely wrong. On a few of my drives I'm getting values which I haven't verified, but on most of them I'm getting 4,294,967,279 / FFFFFFEF which isn't quite a max uint32 unless I'm forgetting something about the math...

[andrewkroh]

I have seen issues with files on a few filesystems. IIRC some file systems report -1 in the stat_t struct when they don't implement this. What file systems were you seeing this problem with?

[DPattee]

I'm testing on two Mac OS X Sierra machines (one is 10.12.5, the other is 10.12.6)

OS Version		s.fs.files		s.fs.mount_point			File System		s.fs.device_name
OS X 10.12.15	4,294,967,279	/						HFS+ Journaled	/dev/disk0s2
OS X 10.12.15	4,294,967,279	/Volumes/<volume2>		HFS+ Journaled	/dev/disk1s2
OS X 10.12.15	4,294,967,279	/Volumes/<volume1>		HFS+ Journaled	/dev/disk2s2
OS X 10.12.15	652				/dev					n/a				devfs
OS X 10.12.15	0				/net					n/a				map -hosts
OS X 10.12.15	0				/home					n/a				map auto_home
OS X 10.12.16	4,294,967,279	/Volumes/Recovery		Apple_Boot		/dev/disk0s3
OS X 10.12.16	4,294,967,279	/						HFS+ Journaled	/dev/disk1
OS X 10.12.16	4,294,967,279	/Volumes/<volume3>		HFS+ Journaled	/dev/disk2s2
OS X 10.12.16	4,294,967,279	/Volumes/Time Machine	HFS+			/dev/disk3s2
OS X 10.12.16	1,205,990,864	/Volumes/<volume4>		NTFS			/dev/disk4s1
OS X 10.12.16	702				/dev					n/a				devfs
OS X 10.12.16	0				/net					n/a				map -hosts
OS X 10.12.16	0				/home					n/a				map auto_home
OS X 10.12.16	1,465,014,458	/Volumes/<volume5>		OSX 10.6.8’s Extended Journaled via AFP	//<user>@<server>._afpovertcp._tcp.local/<volume5>

So, yea, looks like HFS+ is the one that is consistently broken.

Now that I actually make a table of the data, I see a couple other 'things'...

The primary drives on the Mac machines are mounted at /, true. But they do have names that are visible through diskutil list or through an ls /Volumes, so there must be some way to pull the actual name.

Mount points that aren't actually under the /dev/disk* name space or under a /Volumes/ mount point are showing up. I can't think of a case where I'd want to have /dev show up in file system statistics. (And fwiw the /dev entry that claims 702 "files" actually has exactly half that many entries in an ls /dev