gRPC instrumentation - how to add extra context data to spans?

[alertedsnake]

For the gRPC instrumentation plugin, I'm looking to add extra data to spans - in my case I have some metadata information (request ID, user, etc.) available in the request or request context that I'd like to get into the trace somehow.

With OpenTracing, you can provide a SpanDecorator (see code here to the library which allows adding extra information onto a span for a given RPC. Is there a standard way to do that with this library as well?

If not, is it something that might be added in the future?

[alertedsnake]

Ok so I've been digging. I thought about writing another interceptor to add the extra data I need, but that doesn't currently work.

The way this library enables its gRPC interceptor by using a wrapper essentially makes it impossible to add any more gRPC interceptors after the OpenTelemetry interceptor.

If instead this library's interceptor could be added as if it were a standard interceptor (i.e. providing it in the interceptors list argument to grpc.Server()) then it would be easy to add additional instrumetors after. But the wrapper intercept_server() makes this impossible if the instrumetors are not instances of the ABC in grpext within this package.

I've been testing out a rewritten and simplified version of OpenTelemetryServerInterceptor which I've implemented as a standard gRPC interceptor, and then I can control which order things get added, and I can use the active span within a different interceptor, something not possible with the current main branch code.

Is this something that is planned to be addressed?

[codeboten]

hey @alertedsnake, thanks for reporting this issue. Improving the instrumentation support is definitely something we want to work on, that being said the current focus of the main contributors to this repo is to get through the work needed to support the OpenTelemetry api/sdk GA.

If you'd like to tackle this, we'll be happy to review a PR!

[alertedsnake]

I've implemented my own version of most of what I want out of this already, and would be happy to contribute it, but it's going to be a near-complete rewrite, certainly a bit different (see my comments on #1113 also). Are we amenable to that, or are there some guidelines for this?

And I'd have a few questions too - for instance, I don't understand the point of the RpcInfo structure at all.

[codeboten]

I think there's definitely room for improvement, we might even be able to get the original authors of the instrumentation to chime in where you may have questions.