This is because the current span is no longer active at that point. One simple alternate approach is to add a parent span as show below.
from openinference.instrumentation.bedrock import BedrockInstrumentor
from openinference.semconv.trace import SpanAttributes
from opentelemetry.exporter.otlp.proto.grpc.trace_exporter import OTLPSpanExporter
from opentelemetry.sdk.trace import TracerProvider
from opentelemetry.sdk.trace.export import SimpleSpanProcessor
from opentelemetry.trace import Status, StatusCode
endpoint = "http://127.0.0.1:4317"
tracer_provider = TracerProvider()
tracer_provider.add_span_processor(SimpleSpanProcessor(OTLPSpanExporter(endpoint)))
BedrockInstrumentor().instrument(tracer_provider=tracer_provider)
async def event_generator():
input_value = llm_messages[-1].content
with tracer_provider.get_tracer(__name__).start_as_current_span(
"event_generator",
end_on_exit=False,
attributes={
SpanAttributes.INPUT_VALUE: input_value,
SpanAttributes.SESSION_ID: str(chat_id),
SpanAttributes.USER_ID: user.id,
"user_name": user.name,
"org_id": user.org_id,
"org_name": user.org.name,
},
) as span:
chunks = retrieve_reply_for_chat(llm_messages)
response = ""
try:
for chunk in chunks:
response += chunk
yield DataMessage(text=chunk)
except Exception as e:
span.record_exception(e)
span.set_status(Status(StatusCode.ERROR))
logger.exception("Failed to retrieve reply for chat.")
yield FailureMessage(
error_message=(
"Unable to process your request at this time, "
"please try again later. We apologize for the inconvenience."
),
)
else:
span.set_status(Status(StatusCode.OK))
finally:
span.set_attribute(SpanAttributes.OUTPUT_VALUE, response)
span.end()
yield EndMessage()
