Christian Neumüller

> So I think we should avoid any variations for HTTP and have a set of required attributes. Exporters have to deal with breaking up URL into pieces if backend...

> we don't know how user called into HTTP client Why? I'm sure there are cases where we would be wrong, but in general instrumenting the most high-level API will...

Pinging @open-telemetry/python-approvers (and @mariojonke): The suggestion (to which I agree) that the child-most client spans are the most important ones contradicts the current implementation in Python where inner (urllib3) spans...

I think having nested client spans that are marked as client is preferable to having nested client spans where only the innermost one is marked as client. If we want...

> I, unfortunately, have 0 data on how people usually do retries and can only assume things. I assume it's a wild west and high-level logical span is misleading for...

> We need heuristics to tell them which client span represents the best destination You could use resources to determine if these spans come from the same entity. We could...

> I'm not saying there are a lot of users who do a simple loop over high-level API, but they're out there and would come to ask questions and request...

What do I do if I want to encode the HTTP version used in a grpc call? Probably just a HTTP child span under the grpc span?

Related https://github.com/open-telemetry/opentelemetry-specification/issues/1061#issuecomment-704409768 EDIT: But here it may make sense to keep in one issue to define if we even want to support this at all. It seems problematic, both from...

Are there any use cases for arbitrary nesting? I think (multi)maps would be useful to store, e.g., HTTP headers, but what would be the rationale for arbitrary nesting?