Don't set Span.Status for 4xx http status codes for SERVER spans

[iNikem]

Fixes #1818

In addition to all the reasons given in #1818 we have numerous complaints from end-users who are frustrated by seeing all those "error" traces that they can do nothing about.

Changes

Don't set Span.Status for 4xx http status codes for SERVER spans

[Oberon00]

From a theoretical perspective, I'm weakly against this change, but since I don't have user feedback, I yield. 😀

[tigrannajaryan]

I am weakly in favour of the proposed change (4xx not a server error), however I understand that for some people the opposite is desirable. Unfortunately I cannot think of definitive arguments in favour or against either approach. If we cannot reach consensus on this we can make this behavior configurable.

[Oberon00]

If we cannot reach consensus on this we can make this behavior configurable.

That is the one thing I definitely don't want. That configurability should be on the backend.

[carlosalberto]

@iNikem CHANGELOG entry please, as hopefully maintainers can see this upcoming change (and update all their instrumentation) ;)

[tedsuo]

@Oberon00 I disagree about configuration. We need to define a configuration language for OpenTelemetry instrumentation, to handle these issues. But this should be done thoughtfully, as part of the spec. Configuration for instrumentation should not be implemented in an ad-hoc manner; all instrumentation we provide should use the same configuration language.

The instrumentation SIG is working on this configuration issue, but we would greatly appreciate more participation from core Spec members.

As far as this PR goes, I'm fine with the change. My request is that there is some follow up with contrib maintainers and the instrumentation SIG to make sure this change is actually implemented.

[Oberon00]

@Oberon00 I disagree about configuration. We need to define a configuration language for OpenTelemetry instrumentation, to handle these issues.

I'm not against configurability in general, just in this specific case. For HTTP, as discussed here, the span status is just a function of the http.status_code attribute + span kind. That function can be trivially evaluated with a different configuration on the backend to override the status if desired.

[iNikem]

I propose to have a separate discussion about configurability and its location. This PR got enough approvals and enough time has passed for it to be merged. @open-telemetry/technical-committee do you agree?

[carlosalberto]

@iNikem Definitely let's discuss further details on a follow up. Merging, thanks!