view: observable instrument not registered for callback

[utezduyar]

Following runnable code with conditional view is not working and it prints observable instrument not registered for callback message.

package main

import (
	"encoding/json"
	"log"
	"math/rand"
	"os"
	"time"

	"go.opentelemetry.io/contrib/instrumentation/runtime"
	"go.opentelemetry.io/otel/exporters/stdout/stdoutmetric"
	"go.opentelemetry.io/otel/sdk/metric"
	metricsdk "go.opentelemetry.io/otel/sdk/metric"

	"go.opentelemetry.io/otel"
)

func main() {

	// Print with a JSON encoder that indents with two spaces.
	enc := json.NewEncoder(os.Stdout)
	enc.SetIndent("", "  ")
	exporter, err := stdoutmetric.New(
		stdoutmetric.WithEncoder(enc),
		stdoutmetric.WithoutTimestamps(),
	)
	if err != nil {
		panic(err)
	}

	reader := metricsdk.NewPeriodicReader(exporter,
		metricsdk.WithInterval(5*time.Second))

	var drop metric.View = func(i metric.Instrument) (metric.Stream, bool) {
		if i.Name != "process.runtime.go.mem.heap_alloc" {
			return metricsdk.Stream{Aggregation: metricsdk.AggregationDrop{}}, true
		}
		return metricsdk.Stream{}, false
	}

	meterProvider := metricsdk.NewMeterProvider(
		metricsdk.WithReader(reader),
		metricsdk.WithView(drop),
	)
	otel.SetMeterProvider(meterProvider)

	log.Print("Starting runtime instrumentation:")
	err = runtime.Start(runtime.WithMinimumReadMemStatsInterval(time.Second))
	if err != nil {
		panic(err)
	}

	time.Sleep(99 * time.Minute)

}

If I remove the condition from the view and drop the aggregation, it works.

	var drop metric.View = func(i metric.Instrument) (metric.Stream, bool) {
		return metricsdk.Stream{Aggregation: metricsdk.AggregationDrop{}}, true
	}

[utezduyar]

As I wrote on the linked issue: I started to wonder if this is a problem specific to the instrumentation/runtime package. Same way of dropping metrics with a custom View works for other scope like: go.opentelemetry.io/contrib/instrumentation/net/http/otelhttp
Either way depending on the outcome, maybe there needs to be a documentation update on the View API therefore I think the issue is relevant.

[pellared]

When I quickly checked the issue it looks as it is a problem related with observable instruments.

I think this error is produced for any observable instrument with drop aggregation during collection.

[utezduyar]

Isn't this a regression on drop aggregation?
Is there any other way of dropping some of the instruments from the libraries?

[pellared]

Isn't this a regression on drop aggregation?

I do not think it is a regression. It might be a bug that was always there.

Is there any other way of dropping some of the instruments from the libraries?

Nothing comes to my mind apart from creating some metric reader/exporter decorator which would exclude some metric data to be exported... 😬

[scorpionknifes]

I've attempted to draft out a simple fix, not sure if it's good, appreciate any feedback 🙏 #4772

Basically, it's correctly dropping the observable instrument resulting it to be unregistered causing it to fail the registered check