System.Diagnostics.Metrics out of proc monitoring improved histogram support

[CodeBlanch]

Today when listening to metrics via MetricsEventSource histograms are represented as OpenTelemetry summary metrics (more or less).

Summary metrics work well with Prometheus but aren't as well-supported or feature rich as explicit bucket histograms or base2 exponential histograms.

Requesting for .NET 10 we add the ability to opt-into additional histogram types.

Don't have a design yet. Only thoughts...

• Explicit bucket histograms

  • Users should be able to specify custom buckets...
  • Else advice buckets should be respected...
  • Else default buckets should be used.

• Base2 exponential histograms

  • Users should be able to specify MaxSize and MaxScale...
  • Else defaults should be used.

Do we need both? Base2 exponential histogram is the more flexible type as it dynamically adjusts to the values. But it is a newer thing and may not be supported by enough backends to be generally useful.

[rajkumar-rangaraj]

Supporting two types of histograms current default and e2 should be sufficient for .NET 10. The new e2 histogram could be an opt-in. Additional histogram types can be considered in the future based on demand raised through the issue.