prometheus exporter convert instrumentation scope to otel_scope_info metric

[fatsheep9146]

fix #3273
related to #3161

Since #3285 already made an implementation for add target_info metric to prometheus exporter. I think I can wait for this pr merged and add otel_scope_info metric on the basic of his work.

[codecov]

Codecov Report

Attention: Patch coverage is 81.81818% with 8 lines in your changes missing coverage. Please review.

Project coverage is 77.8%. Comparing base (49b62ae) to head (7efab63).
Report is 1849 commits behind head on main.

Files with missing lines	Patch %	Lines
exporters/prometheus/exporter.go	79.4%	7 Missing and 1 partial ⚠️

Additional details and impacted files

@@          Coverage Diff          @@
##            main   #3357   +/-   ##
=====================================
  Coverage   77.7%   77.8%           
=====================================
  Files        164     164           
  Lines      11457   11488   +31     
=====================================
+ Hits        8913    8942   +29     
- Misses      2342    2343    +1     
- Partials     202     203    +1     

Files with missing lines	Coverage Δ
exporters/prometheus/config.go	100.0% <100.0%> (ø)
exporters/prometheus/exporter.go	82.5% <79.4%> (+1.4%)	⬆️

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[dashpole]

Note that this does not fully fix #3161. I believe you also need to implement open-telemetry/opentelemetry-specification#2890

[fatsheep9146]

Note that this does not fully fix #3161. I believe you also need to implement open-telemetry/opentelemetry-specification#2890

You are right, I mock a test as following.

If use two different meter to create two metric with same name and enable scope info, no errors happen

func TestMetricWithSameName(t *testing.T) {
	exporter, err := New()
	assert.NoError(t, err)

	provider := metric.NewMeterProvider(
		metric.WithReader(exporter),
	)

	httpCounter, err := provider.Meter("http").SyncInt64().Counter("error_count", instrument.WithUnit(unit.Dimensionless))
	assert.NoError(t, err)
	httpCounter.Add(context.TODO(), 1)

	sqlCounter, err := provider.Meter("sql").SyncInt64().Counter("error_count", instrument.WithUnit(unit.Dimensionless))
	assert.NoError(t, err)
	sqlCounter.Add(context.TODO(), 1)

	t.Logf("serving metrics at localhost:2223/metrics")
	http.Handle("/metrics", promhttp.Handler())
	err = http.ListenAndServe(":2223", nil)
	if err != nil {
		t.Fatalf("error serving http: %v", err)
		return
	}
}

curl metric port will produce following metric

# HELP error_count_ratio_total
# TYPE error_count_ratio_total counter
error_count_ratio_total{otel_scope_name="http",otel_scope_version=""} 1
error_count_ratio_total{otel_scope_name="sql",otel_scope_version=""} 1

But if I disable scope info like this

func TestMetricWithSameName(t *testing.T) {
	exporter, err := New(WithoutScopeInfo())
	... // following codes keep unchanged
}

the error happned

An error has occurred while serving metrics:

collected metric "error_count_ratio_total" { counter:<value:1 > } was collected before with the same name and label values

I think I can solve this problem in another pr? Is that ok ? @dashpole @MrAlias

[dashpole]

If use two different meter to create two metric with same name and enable scope info, no errors happen

Can you try with a different description on each? I think that is why you aren't reproducing the issue.

Can you also try with different Counter vs UpDownCounter?

[fatsheep9146]

Can you try with a different description on each? I think that is why you aren't reproducing the issue.

For two Counter of two different meter with same name, different Description, an error happend @dashpole

code

...
httpCounter, err := provider.Meter("http").
  SyncInt64().Counter(
  "error_count",
  instrument.WithUnit(unit.Dimensionless),
  instrument.WithDescription("http error count"))
  httpCounter.Add(context.TODO(), 1)

sqlCounter, err := provider.Meter("sql").
  SyncInt64().Counter(
  "error_count",
  instrument.WithUnit(unit.Dimensionless),
  instrument.WithDescription("sql error count"))
sqlCounter.Add(context.TODO(), 1)
 ...

error

collected metric error_count_ratio_total label:<name:"otel_scope_name" value:"sql" > label:<name:"otel_scope_version" value:"" > counter:<value:1 >  has help "sql error count" but should have "http error count"

[fatsheep9146]

Can you also try with different Counter vs UpDownCounter?

For one Counter and another UpDownCounter of two different meter with same name, same description, no error happend @dashpole

code

httpCounter, err := provider.Meter("http").
  SyncInt64().Counter(
  "error_count",
  instrument.WithUnit(unit.Dimensionless))
httpCounter.Add(context.TODO(), 1)

sqlCounter, err := provider.Meter("sql").
  SyncInt64().UpDownCounter(
  "error_count",
  instrument.WithUnit(unit.Dimensionless))
sqlCounter.Add(context.TODO(), 1)

result

# HELP error_count_ratio
# TYPE error_count_ratio gauge
error_count_ratio{otel_scope_name="sql",otel_scope_version=""} 1
# HELP error_count_ratio_total
# TYPE error_count_ratio_total counter
error_count_ratio_total{otel_scope_name="http",otel_scope_version=""} 1
...

[MrAlias]

Note that this does not fully fix #3161. I believe you also need to implement open-telemetry/opentelemetry-specification#2890

You are right, I mock a test as following.

If use two different meter to create two metric with same name and enable scope info, no errors happen

func TestMetricWithSameName(t *testing.T) {
	exporter, err := New()
	assert.NoError(t, err)

	provider := metric.NewMeterProvider(
		metric.WithReader(exporter),
	)

	httpCounter, err := provider.Meter("http").SyncInt64().Counter("error_count", instrument.WithUnit(unit.Dimensionless))
	assert.NoError(t, err)
	httpCounter.Add(context.TODO(), 1)

	sqlCounter, err := provider.Meter("sql").SyncInt64().Counter("error_count", instrument.WithUnit(unit.Dimensionless))
	assert.NoError(t, err)
	sqlCounter.Add(context.TODO(), 1)

	t.Logf("serving metrics at localhost:2223/metrics")
	http.Handle("/metrics", promhttp.Handler())
	err = http.ListenAndServe(":2223", nil)
	if err != nil {
		t.Fatalf("error serving http: %v", err)
		return
	}
}

curl metric port will produce following metric

# HELP error_count_ratio_total
# TYPE error_count_ratio_total counter
error_count_ratio_total{otel_scope_name="http",otel_scope_version=""} 1
error_count_ratio_total{otel_scope_name="sql",otel_scope_version=""} 1

But if I disable scope info like this

func TestMetricWithSameName(t *testing.T) {
	exporter, err := New(WithoutScopeInfo())
	... // following codes keep unchanged
}

the error happned

An error has occurred while serving metrics:

collected metric "error_count_ratio_total" { counter:<value:1 > } was collected before with the same name and label values

I think I can solve this problem in another pr? Is that ok ? @dashpole @MrAlias

@dashpole what's the expected behavior here? Just drop the second attempt to collect the second instrument and log an error?

[dashpole]

what's the expected behavior here? Just drop the second attempt to collect the second instrument and log an error?

From open-telemetry/opentelemetry-specification#2890

Example approach

• Keep a map of metric name -> Description when constructing a batch of metrics for a scrape.
• Each time a new metric (with Description) is added, look up the name in the map:
  • If it isn't found, add it to the map
  • If it is found, and the desc is the same, do nothing
  • If it is found, and has conflicting type information, drop the metric + log a warning
  • If it is found, and has conflicting help or unit, use the stored help + unit instead of the provided one + log a warning.

Basically, we need the description of all metrics with the same name to match.

[MrAlias]

what's the expected behavior here? Just drop the second attempt to collect the second instrument and log an error?

From open-telemetry/opentelemetry-specification#2890

Example approach

• Keep a map of metric name -> Description when constructing a batch of metrics for a scrape.

• Each time a new metric (with Description) is added, look up the name in the map:

  • If it isn't found, add it to the map
  • If it is found, and the desc is the same, do nothing
  • If it is found, and has conflicting type information, drop the metric + log a warning
  • If it is found, and has conflicting help or unit, use the stored help + unit instead of the provided one + log a warning.

Basically, we need the description of all metrics with the same name to match.

👍

@dashpole are you good with merging this and then adding that tracking in another PR? That seems to make sense to me.

[dashpole]

@dashpole are you good with merging this and then adding that tracking in another PR? That seems to make sense to me.

Yes.