Tracing async generator pipelines: where does the context actually break?
We're running async Python generators that chain through 3-4 microservices. OpenTelemetry traces show gaps — the context seems to drop when generators yield across HTTP boundaries. We tried propagating traceparent manually but it's fragile. How are you handling distributed tracing when the work unit is an async generator, not a simple request/response? Frameworks tried: otel-sdk-python with manual context attach/detach. Curious if anyone solved this at the framework level vs. patching each yield point.