Support forwarding OpenCensus spans over TTRPC for shim task service

[Kevin Parsons (kevpar)]

No description provided.

[Justin (jterry75)]

LGTM. I'm only wondering about duplicate error logs but its not a huge deal

[Kevin Parsons (kevpar)]

Justin (@jterry75) Assuming you mean duplicate error spans between the ones created by octtrpc and the "implementation side" span, my stance is we should adapt our queries to work with it. I think mostly from a query perspective we should be looking at errors in "user facing" spans (e.g. CRI entry points), so this shouldn't be an issue there. For the case of looking to see what are the top errors across all spans, we will have to deal with the duplicate, but I don't think it will be a big deal.

The behavior here also matches what is done with ocgrpc, so I think we'll need to deal with it even if we tried to do something else in octtrpc.

I'm going to check this in, but happy to discuss this further next week.

[Justin (jterry75)]

Here is what I am wondering in a nutshell:

Here is what we do today:

func (s *svc) SomeMethod(ctx context.Context, p *Parameters) (r *Return, err error) {
    ctx, span := trace.StartSpan(ctx, "SomeMethod)
    span.AddAttributes(...) // stuff with `p`
    defer func() {
        if err == nil {
            span.AddAttributes(...) // stuf with `r`
        }
    }()
    // Method impl
}

But since we now create the entry spans automatically I was wondering if an entry method should now do:

func (s *svc) SomeMethod(ctx context.Context, p *Parameters) (r *Return, err error) {
    span := trace.FromContext(ctx)
    if span != nil {
        span.AddAttributes(...) // stuff with `p`
        defer func() {
            if err == nil {
                span.AddAttributes(...) // stuff with `r`
            }
        }()
    }
    // Method impl
}

So instead of overwriting the entry span with an immediate child we instead enlighten the attributes of an entry span.