Open william-tran opened 4 years ago
@dmitryax Does this issue reflect some of your concern in https://github.com/open-telemetry/opentelemetry-collector/pull/844?
Talked to @carlosalberto, looks like this is a sub-issue of #114
And #65 is a sub-issue of this. EDIT: Or is this issue a duplicate of #65?
Note from the maintainers meeting, @carlosalberto committed to completing this by the end of the week, which we need to do to hit our tracing spec RC deadlines.
from the issue triage meeting today with TC, allowed for GA for editorial change only
In the Java shim, the first reference is used as the parent regardless of the reference type: https://github.com/open-telemetry/opentelemetry-java/blob/v0.3.0/opentracing_shim/src/main/java/io/opentelemetry/opentracingshim/SpanBuilderShim.java#L74-L89
In Golang, its the first
CHILD_OF
: https://github.com/open-telemetry/opentelemetry-go/blob/v0.4.2/bridge/opentracing/bridge.go#L540-L548In the Jaeger Java client, when spans are converted to thrift, the parent is set if and only if there is a single child reference: https://github.com/jaegertracing/jaeger-client-java/blob/v1.2.0/jaeger-thrift/src/main/java/io/jaegertracing/thrift/internal/reporters/protocols/JaegerThriftSpanConverter.java#L50. I'm currently hitting this using the thrift HTTP receiver in the opentelemetry-collector. See https://github.com/jaegertracing/jaeger-client-java/issues/705 for tracking this behaviour.
The spec should explicitly call out what the translation should be to avoid inconsistencies when translating to OTel.