Configuring Asynchronous Tracing with HTTP in Tokio Runtime - SpanExporter with_tonic vs with_http

[antonmithun-work]

I've read that gRPC is faster than HTTP, but I have a question about configuring the trace provider with HTTP. Do we need to specify anything specific to ensure that tracing tasks run asynchronously? Specifically, does the following configuration for with_http work asynchronously, or does it operate in a slower synchronous manner? And do we need to explicitly configure it to run within the Tokio runtime on separate threads?

 let exporter = SpanExporter::builder()
            //.with_tonic()
            .with_http()
            .with_endpoint(&config.otlp_trace_url)
            .build().unwrap();

        let tracer_provider = opentelemetry_sdk::trace::SdkTracerProvider::builder()
            .with_batch_exporter(exporter)
            .with_resource(
                Resource::builder_empty()
                    .with_service_name(config.service_name.clone())
                    .build()
            )
            .build();

[cijothomas]

The default behavior of with_batch_exporter is that it spawns own dedicated thread to handle exports. i.e it is never done in the user-threads.
There is experimental feature to instead use one of the async runtime tasks instead of own threads, but this is usually useful if you are in an environment where thread creation is not possible.