roadmap to make statsHandler stable

[sahejsingh]

https://github.com/grpc/grpc-go/blob/master/stats/stats.go#L21

// All APIs are experimental.

Reading the above gives an impression that the statsHandler api is not stable. Does this mean

• api can be removed?
• statsHandler interface or the stats it offers can change?

If answer to above is yes, is there a plan by when can we expect it to become stable?
If answer is no, can you please explain why that line exists? how should we interpret that.

[dfawley]

Does this mean

• api can be removed?
• statsHandler interface or the stats it offers can change?

Yes, this is the idea.

is there a plan by when can we expect it to become stable?

I am going to be making some changes here within in the next 2-3 weeks to expose RPC retry information; hopefully in a non-breaking way, but in the process of doing that, I will assess the possibility of stabilizing these APIs as well.

[dfawley]

Update: I think the current basic API here can be marked as stable, with the possible exception of: can we remove some of the specific fields that are deprecated or unnecessary first? E.g. End.Trailer, OutTrailer.WireLength, and all the time.Time fields.

[pellared]

@dfawley, are there any plans to make the package stable in near time? We would like to have move confidence on the stability of the stats package for OpenTelemetry instrumentation in order favor it over interceptors API.

Related issue: open-telemetry/opentelemetry-go-contrib#4319

Related PR: open-telemetry/opentelemetry-go-contrib#5196

[dfawley]

@pellared

I'm not sure if you're aware, but gRPC-Go intends to publish our own OTel instrumentation (using the stats package, which we can do safely since we control both sides of the API). The still-in-progress work is coming together in https://github.com/grpc/grpc-go/tree/master/stats/opentelemetry

Regarding the stability, there are a few changes coming - e.g. we will soon be removing the Data field from in/out payload (marked deprecated via #7121). There may be other changes needed as well, but we will try our best to have migration strategies where possible.

[pellared]

I'm not sure if you're aware, but gRPC-Go intends to publish our own OTel instrumentation

Awesome 👍

Does it mean that we after this is done we could deprecate https://pkg.go.dev/go.opentelemetry.io/contrib/instrumentation/google.golang.org/grpc/otelgrpc?

Do you intend to follow https://opentelemetry.io/docs/specs/semconv/rpc/grpc/? Are you involved in stabilizing the OpenTelemetry Semantic Conventions for RPC and gRPC?

[dfawley]

Does it mean that we after this is done we could deprecate

We hope you will! :)

Do you intend to follow opentelemetry.io/docs/specs/semconv/rpc/grpc?

TBH I don't know. 😆 We are following our cross-language design here: https://github.com/grpc/proposal/blob/master/A66-otel-stats.md

Maybe @yashykt knows about both and can say off-hand whether they're the same or different.

[dashpole]

open-telemetry/semantic-conventions#627 was the latest change in that area. I believe OTel's migration plan to the gRPC metrics is still TBD.