[processor/spanmetricsprocessor] Add ms unit to latency buckets

[Yamakaky]

Description: Define the latency metric unit as milliseconds for spanmetricsprocessor.

Testing: unit tests

Documentation: the metric was already documented as being in milliseconds.

[linux-foundation-easycla]

The committers listed above are authorized under a signed CLA.

• ✅ login: Yamakaky / name: Yamakaky (4ad768f6d94b368151dd10ecd86597620e84654b)

[dmitryax]

LGTM. cc @albertteoh as code owner

[Yamakaky]

Should I update the changelog?

[dmitryax]

Yes, I believe it should be mentioned in the changelog. please create an enhancement entry in unreleased directory

[albertteoh]

LGTM.

Forgive my ignorance, but what are the use cases for the optional metric units in the telemetry data itself? My understanding, at least for prometheus, is that units are not stored with the metric data but, instead, are carried in the metric name itself like http_request_duration_seconds.

[Yamakaky]

Hum, relevant references:

• https://github.com/open-telemetry/opentelemetry-specification/blob/main/specification/metrics/semantic_conventions/README.md#units
• https://prometheus.io/docs/practices/naming/

Based on that, I think the metric should be renamed to something like span.duration and potentially use seconds as units. What do you think?

Also, the unit is part of the data model https://opentelemetry.io/docs/reference/specification/metrics/datamodel/#opentelemetry-protocol-data-model. I don't know if many vendors use it, but it would allow auto unit discovery.