Prometheus metric names incorrectly append unit

[JamesNK]

Bug Report

List of all OpenTelemetry NuGet
packages and version that you are
using (e.g. OpenTelemetry 1.0.2):

• OpenTelemetry.Exporter.Prometheus.AspNetCore 1.6.0-alpha.1
• OpenTelemetry.Extensions.Hosting 1.6.0-alpha.1

Runtime version (e.g. net462, net48, netcoreapp3.1, net6.0 etc. You can
find this information from the *.csproj file):

• net7.0

Symptom

The .NET prometheus exporter doesn't correctly follow the specification for outputting units:

https://github.com/open-telemetry/opentelemetry-specification/blob/065b25024549120800da7cda6ccd9717658ff0df/specification/compatibility/prometheus_and_openmetrics.md?plain=1#L235-L240

For example, a unit in brackets is included in the output. Brackets should be removed.

.NET code:

Output:

This counter shouldn't have a unit because {TestUnitWithBrackets} is surrounded by brackets.

What is the expected behavior?

No unit on the counter.

What is the actual behavior?

The counter has a unit of TestUnitWithBrackets

Reproduce

using OpenTelemetry.Metrics;
using System.Diagnostics.Metrics;

var builder = WebApplication.CreateBuilder(args);
builder.Services.AddOpenTelemetry()
    .WithMetrics(builder =>
    {
        builder.AddPrometheusExporter();

        builder.AddMeter("HelloWorld");
    });

var app = builder.Build();

var meter = new Meter("HelloWorld");
var counter = meter.CreateCounter<int>("Test.Counter", unit: "{TestUnitWithBrackets}");

app.MapGet("/", () =>
{
    counter.Add(1);
    return "Hello World!";
});
app.MapPrometheusScrapingEndpoint();

app.Run();

We will close this issue if:

• The repro project you share with us is complex. We can't investigate custom
  projects, so don't point us to such, please.
• If we can not reproduce the behavior you're reporting.

Additional Context

Add any other context about the problem here.

[JamesNK]

There are other things the .NET exporter isn't doing. It's supposed to map to Prometheus units. For example, s becomes seconds.

For example, this is what go has to implement this behavior: https://github.com/open-telemetry/opentelemetry-collector-contrib/blob/9a9d4778bbbf242dba233db28e2fbcfda3416959/pkg/translator/prometheus/normalize_name.go#L30

Here is Java implementing this feature: open-telemetry/opentelemetry-java#5400

[JamesNK]

I think it's important that this is fixed sooner rather than later. When it is fixed the metric names in Prometheus output will change, and people will have to react to the new names. For example, people who have written dashboards or alerts on the names will be broken and need to update Prometheus/Grafana/etc.

The sooner it is fixed, the fewer people will be impacted.