It turned out that linked spans cannot easily be extracted by the
functions that the SDK currently offers, e.g. SpanContext does
not even contain the ID of the parent span.
And the logic that was implemented up until now was doing it all
wrong and ended-up linking each trace ID that it parsed back to
itself (see #69).
With the need for smarter linking in mind (#70), it appears that
actually the right thing to do is to keep generating new traces
for HTTP flows, however link to existing spans whenever tracing
headers are present.
It turned out that linked spans cannot easily be extracted by the functions that the SDK currently offers, e.g.
SpanContext
does not even contain the ID of the parent span. And the logic that was implemented up until now was doing it all wrong and ended-up linking each trace ID that it parsed back to itself (see #69). With the need for smarter linking in mind (#70), it appears that actually the right thing to do is to keep generating new traces for HTTP flows, however link to existing spans whenever tracing headers are present.Fixes #69 Fixes #70