Open avilagaston9 opened 3 days ago
IMO, we should have two traces: one for the Batcher and another for the Aggregator. This approach removes the dependency between them and also enables the possibility of registering the Aggregator spans when running more than one Batcher simultaneously.
When the Batcher cancels a batch, we don't finalize the trace in case the batch has already been sent and the Aggregator receives it. However, if the batch was actually canceled, the trace metadata will remain in the store indefinitely.