Custom Propagator not working

[facucus]

Description

We want to create a new propagator to fetch traceid/spanid from a URL, I create a new propagator and assign it to the instrumentation but I never see my code being executed.

Steps to reproduce

class UrlPropagator  {
    fields() {
        return ['traceid', 'spanid'];
    }

    inject(context, carrier: unknown, setter: TextMapSetter): void {
    }

    extract(context, carrier, getter) {
        const traceId = getter.get(carrier, 'traceId');
        const spanId = getter.get(carrier, 'spanId');
        console.log("extract: ", traceId, spanId)
        if (!traceId || !spanId) {
            return context;
        }
        const spanContext = {
            traceId,
            spanId,
            traceFlags: TraceFlags.SAMPLED,
        };
        console.log("spanContext: ", spanContext)
        return trace.setSpanContext(context, spanContext);
    }
}

...initializeFaro
new TracingInstrumentation({
      propagator: new UrlPropagator()
}),

Expected behavior

We expect the propagator be executed and at least see the console.log

Actual behavior

The propagator it's never executed

[cedricziel]

I think the Web SDK is behaving correctly in this case.

The intention of a propagator is to extract information from an incoming request and to inject information into an outgoing request. OpenTelemetry tracing propagates context forward, not backwards.

Ref: https://opentelemetry.io/docs/specs/otel/context/api-propagators/#extract

Extracts the value from an incoming request. For example, from the headers of an HTTP request.

In case the intention is to capture every interaction in a single trace, the OpenTelemetry docs list an example how to map a server-started trace and use it as root-span for your browser level tracing needs. https://opentelemetry.io/docs/instrumentation/js/getting-started/browser/#example-application

We are exploring ways to connect client side events stronger with server side context by using a lose coupling. We are tracking #216 which captures the initial ideation around using server-timing http headers and enhancements to our new fetch instrumentation to create the context necessary.

[codecapitano]

@facucus @cedricziel Can we close this issue or is anything left to clarify?