Fixing formatting issue with :infinity

[akoutmos]

Came across this error today in an app running PromEx:

[error] Task #PID<0.4021.0> started from Whoosh.PromEx.ETSCronFlusher terminating
** (FunctionClauseError) no function clause matching in Peep.Prometheus.format_value/1
    (peep 3.2.0) lib/peep/prometheus.ex:116: Peep.Prometheus.format_value(:infinity)
    (peep 3.2.0) lib/peep/prometheus.ex:93: anonymous fn/2 in Peep.Prometheus.format_standard/2
    ...

And while I am not 100% sure as to how Peep works internally or how :infinity ended up in that value, according to the Prometheus spec +Inf, -Inf, and NaN are all valid values for a metric. This PR just adds support for formatting :infinity -> +Inf values to solve my immediate issue...but not sure what to do about the other two.

[rkallos]

Thank you for your contribution!

If you ever figure it out, I'd be interested to know how :infinity wound up in Peep.

not sure what to do about the other two.

According to this, Erlang will raise a badarith exception instead of returning +Inf, -Inf, or NaN. Elixir inherits this behavior. For now, I think it's fine to avoid implementing those.

It looks like your contribution will fix your issue, but I'm wondering if your issue isn't a symptom of a larger issue. Peep seems to support associating and storing arbitrary Erlang terms with a Telemetry.Metrics.LastValue, which is different from counter, sum, and distribution metrics, which, for one reason or another, only support integers and sometimes floats.

Looking at other libraries for inspiration, I found:

• TelemetryMetricsPrometheus.Core appears to drop non-numeric measurement values.
• TelemetryMetrics doesn't outright specify that metrics values should only be numbers, but it's worth noting that the conversion between units only works on numeric values.
• :telemetry's typespecs suggest that measurements can either be scalar numbers, or a map(), though the documentation does not specify a type for keys or values within a map.

It's my feeling that it would be better to remove support altogether for non-numeric values being associated with Telemetry.Metrics.LastValue metrics, rather than to add the special-case where the atom :infinity is handled appropriately.

What are your thoughts?