Felix Barnsteiner

There are also situations that look like the trace root is in another system whereas the root is just coming in late or is lost. Having the non-root traces appear...

I think a relatively easy fix for that would be to introduce a flag that allows to disregard the sampling decision from the incoming traceparent header. This would have to...

> @felixbarny that would be an unfortunate solution as it would allow you to either have the problems mentioned above or not have distributed tracing - we do rely on...

That's right, good point. In your infra, are all public requests routed trough a dedicated load balancer? If so, could you just drop the `traceparent` header there?

Seems there are lots of different use cases that depend on the specifics of each particular setup. There's no one-size-fits-all solution. The different dimensions in determining whether to continue a...

> any consumers of the API that happen to use elastic APM as well will be sending us valid headers without even knowing about it Yeah, good point. The request...

I think we can simplify the `trace_continuation_strategy` when adding support for span links. Instead of differentiating between `restart` and `resample`, we would only have `restart` but also add a span...

> @felixbarny project.name is an APM agent config option? (A new name for what you called deployment_name above?) Yes. I was trying out `project_name` vs `deployment_name` but meant the same...

No concrete plans currently. However, you could try using the OpenTelemetry C++ SDK. See also https://discuss.elastic.co/t/instrumenting-a-c-application/245026 for a similar question in the forum.

Yep, that would lead to duplicated metrics in different indices. It probably makes sense for agents to offer an option to disable Prometheus metric collection. Not sure if we need...