Should celery add a span link instead?

[Kludex]

Describe your environment

Currently, the celery task that is executing gets the context from the headers, and attaches it to the new span:

opentelemetry-python-contrib/instrumentation/opentelemetry-instrumentation-celery/src/opentelemetry/instrumentation/celery/__init__.py

Lines 173 to 180 in e3ba54b

	token = context_api.attach(tracectx) if tracectx is not None else None
	logger.debug("prerun signal start task_id=%s", task_id)
	operation_name = f"{_TASK_RUN}/{task.name}"
	span = self._tracer.start_span(
	operation_name, context=tracectx, kind=trace.SpanKind.CONSUMER
	)

I don't think this is the correct behavior. It should actually be a span link - according to the docs:

Links exist so that you can associate one span with one or more spans, implying a causal relationship. For example, let’s say we have a distributed system where some operations are tracked by a trace.

Which is exactly this case.

When a web framework triggers a celery task, it shouldn't be on the same trace. It's already a new one... But they are linked through the span link.

Would you like to implement a fix?

Yes

[samuelcolvin]

agreed, I think this is exactly what span links are for.

[unflxw]

To add to the agreement, note that this is the behaviour speciified in the OpenTelemetry Semantic Conventions for Messaging Spans (see "Trace structure" and "Examples")