fix(tracing/core): sampled flags and propagation

[samugi]

Summary

fix an issue where incoming propagation headers with disabled sampling (sampled flag = 0) produced invalid propagation of noop spans and broken traces

This fix:

• ensures propagation is not attempted on noop spans
• ensures no span is sampled when sampling is disabled in the incoming headers
• propagates the correct header when sampling is disabled in the incoming headers

Checklist

• The Pull Request has tests
• [not needed] There's an entry in the CHANGELOG
• [not needed] There is a user-facing docs PR against https://github.com/Kong/docs.konghq.com - PUT DOCS PR HERE

Full changelog

• [Implement ...]

Issue reference

https://konghq.atlassian.net/browse/KAG-1184
https://konghq.atlassian.net/browse/KAG-1176

[StarlightIbuki]

I have a question after reviewing the code (though not related to the fix). If every node always inherits "should_sample", and none of the upstream nodes will ever send the trace, what's the point we propagate the tracing context via headers?

[samugi]

I have a question after reviewing the code (though not related to the fix). If every node always inherits "should_sample", and none of the upstream nodes will ever send the trace, what's the point we propagate the tracing context via headers?

I think the answer is in here, for example: A component may also fall back to probability sampling and set the sampled flag to 1 for the subset of requests.. In other words we don't know whether upstream will use Deferred sampling or some other strategy so the right thing to do would be propagating the context.