Add system metrics reporting total memory capacity or clarify how to recover existing ones

[mx-psi]

trafficstars

Current system metrics cover usage of memory, paging/swap memory and filesystems, but we don't currently support the total capacity of any of these systems as a separate metric.

The following metric can't be recovered from the existing system metrics and would need to be added to support this:

Name	Description	Units	Instrument Type	Value Type	Attribute Key	Attribute Values
system.disk.limit	Total memory available in the disk.	By	UpDownCounter	Int64	device	(identifier)

The following metric may be recovered from system.memory.usage, but the current description of the attribute values of state is insufficient to recover this.

Name	Description	Units	Instrument Type	Value Type	Attribute Key	Attribute Values
system.memory.limit	Total memory available in the machine. Does not include paging/swap memory.	By	UpDownCounter	Int64	n/a	n/a

The following metrics are available as the sum of used and free (and reserved for the file system one). They could be added as a convenience metric:

Name	Description	Units	Instrument Type	Value Type	Attribute Key	Attribute Values
system.filesystem.limit	Total memory available in the disk.	By	UpDownCounter	Int64	device	(identifier)
					state	used, free, reserved
					type	ext4, tmpfs, etc.
					mode	rw, ro, etc.
					mountpoint	(path)
system.paging.limit	Total paging/swap memory available.	By	UpDownCounter	Int64	n/a	n/a

Items for this issue:

• [ ] Add system.disk.limit to the specification
• [x] Clarify set of attribute values for system.memory.usage state and consider using system.memory.total
• [ ] Consider adding system.filesystem.limit and system.paging.total if convenience justifies it

This would be part of open-telemetry/opentelemetry-specification/issues/3556 if approved.

[trask]

check out https://github.com/open-telemetry/semantic-conventions/blob/main/docs/general/metrics.md#do-not-use-total

.limit is probably the closest existing convention: https://github.com/open-telemetry/semantic-conventions/blob/main/docs/general/metrics.md#instrument-naming

[mx-psi]

Thanks, I changed all .total suffixes by .limit suffixes above :)

[yevgentrukhin]

Curious for next steps, will there be a PR for this change to semantic convention?

[mx-psi]

Curious for next steps, will there be a PR for this change to semantic convention?

@yevgentrukhin yes, this is part of the system semantic conventions WG roadmap and will be addressed before stabilization of system metrics. If you are interested in having this sooner, I am happy to review PRs related to this and you can join our weekly meeting to discuss if necessary (see here for details)

[mx-psi]

Clarify set of attribute values for system.memory.usage state and consider using system.memory.total

I have marked this one as done, since #89 we have the the total value for state which would be system.memory.limit

[mx-psi]

Discussed on January 18th System Semantic Conventions WG meeting, we don't consider this one to be a blocker for system metrics GA unless the new metrics affect existing metrics. @mx-psi to double check this.

[rogercoll]

@joaopgrassi Sorry, I think in the https://github.com/open-telemetry/semantic-conventions/pull/1356 PR I did not specify that it was partially solving this issue. I reckon system.filesystem.limit and system.paging.limit metrics still need to be added.