FaaS metrics - Should we capture metrics for clients invoking functions?

[joaopgrassi]

In #88 (comment) we discussed and aligned that all existing FaaS metrics are related to the FaaS itself and not by whoever is invoking them (clients).

So, the conventions were modified to only include metrics related to the FaaS itself, and not by clients invoking them.

We need to discuss if:

• We want/need to record metrics that measure things related to clients invoking a function (how long it took, how many errors etc)
  • If so, we should decide how to name them. A proposal was to introduce them under a new namespace faas.client.*
• We don't need them and are fine with only the "faas-side" metrics only
  • As @Oberon00 then we may also need to remove them from tracing Generate FaaS metrics semconv from YAML #88 (comment)

[jviau]

This would not apply to Azure functions as there is not concept of a FaaS client. All calls to Azure Functions are done through existing clients: http, messaging, etc. Do other FaaS implementations have a specific client? Do we have a common enough understanding of what a FaaS client is to make this part of the core spec?

As we are talking about naming these with a scheme of faas.client.*, should we also name the non-client spans with a differentiator as well? faas.server.*, faas.host.*, or faas.worker.*? Or something else.