libcni: add OpenTelemetry trace context propagation

[jipperinbham]

This enables distributed tracing support by passing span identifiers from the caller to CNI plugins through environment variables. When a valid span context exists, it is propagated using the TRACEPARENT, TRACESTATE, and BAGGAGE environment variables following the OpenTelemetry specification OTEP 0258.

This allows CNI plugins to participate in distributed traces, helping with performance analysis and debugging across the plugin execution lifecycle. I does not create any spans and instead leaves that responsibility to the caller and CNI plugins. Below is an example for extracting the trace context via environment variables:

func extractTraceparent(ctx context.Context) context.Context {
	traceParent, ok := os.LookupEnv("TRACEPARENT")
	if !ok {
		return ctx
	}

	traceParts := strings.Split(traceParent, "-")
	if len(traceParts) != 4 {
		return ctx
	}

	traceID, err := trace.TraceIDFromHex(traceParts[1])
	if err != nil {
		return ctx
	}

	spanID, err := trace.SpanIDFromHex(traceParts[2])
	if err != nil {
		return ctx
	}

	flagBytes, err := hex.DecodeString(traceParts[3])
	if err != nil {
		return ctx
	}

	var ts trace.TraceState
	if traceState, ok := os.LookupEnv("TRACESTATE"); ok {
		if ts, err = trace.ParseTraceState(traceState); err != nil {
			// no-op
		}
	}

	// traceBaggage, err := baggage.Parse(os.Getenv("BAGGAGE"))
	// if err != nil {
	// 	slog.With("error", err.Error()).Error(fmt.Sprintf("Failed to parse baggage %s", err.Error()))
	// }

	scc := trace.SpanContextConfig{
		TraceID:    traceID,
		SpanID:     spanID,
		TraceFlags: trace.TraceFlags(flagBytes[0]),
		TraceState: ts,
		Remote:     true,
	}

	return trace.ContextWithSpanContext(ctx, trace.NewSpanContext(scc))
}

Fixes #561

[jipperinbham]

I'm guessing this is causing the CI failure. @squeed would you prefer I rollback to 1.21?

[MikeZappa87]

@jipperinbham can you fix the linter and we can move this forward?

[jipperinbham]

@MikeZappa87 thanks for the ping! I've updated this work and containerd/go-cni#132 to having passing CI.

[coveralls]

coverage: 63.012% (-0.3%) from 63.325% — jipperinbham:feat/otel-tracing into containernetworking:main