How to use parent-less spans for 'additional reactor work' when coupled with agent-based instrumentation

[crankydillo]

I'm working on a reactor-based web server. For each inbound request, I need an 'official' trace that correlates it with all outbound HTTP requests for 'official work'. Auto-instrumentation seems to work quite well for this. However, I also need to make additional outbound HTTP requests that need traces, but should not be associated with the 'official trace'. We can call it the 'shadow trace' for 'shadow work'.

I've read as much as I could find on open-telemetry and reactor and have the following example code.

Mono<String> process(Request inbound) {
   Span shadowSpan = createSpan("shadow"); // do all the no-parent stuff
   try (Scope shadowScope = shadowSpan.makeCurrent()) {
      shadowWork()
            .doFinally(sig -> shadowSpan.complete())
            .subscribe();
   }
   return officialWork();
}

Of course, 'shadow work' may make HTTP calls that happen long after the 'official work' completes.
In all the scenarios I've thrown at it, this seems to work; however, I have also been burned by 'testing' my way to conclusions about how reactor works.

Does the above code look good for what I've described when paired with auto-instrumentation? If not, do you have any guidance? I'm thinking that TraceSubscriber is the secret sauce that makes things works.