[Bug]: Plugin generates performance data with value Inf or -Inf

[RincewindsHat]

Please try to fill out as much of the information below as you can. Thank you!

• Yes, I've searched similar issues on GitHub and didn't find any.

Which version contains the bug?

1.0.2

Describe the bug

On some occasions (don't know which yet), the plugin produces performance data with the value Inf or -Inf.
The mode was query with, appropriately, a custom query.
Without investigation I am assuming that the retrieved value is empty or invalid in a different way.

As for a solution, I would suggest to filter out such values as I cannot imagine that to be a valid measurement of anything.

How to recreate the bug?

No response

[RincewindsHat]

ref/NC/820245

[RincewindsHat]

Might also be something to provide in go-check

[martialblog]

Yeah I think we should fix that in the library.

[martialblog]

Recreated the issue in a unittest: https://github.com/NETWAYS/go-check/tree/fix/perfdata-inf

Should me move the issue to the go-check library?

[RincewindsHat]

Recreated the issue in a unittest: https://github.com/NETWAYS/go-check/tree/fix/perfdata-inf

Should me move the issue to the go-check library?

yes

[martialblog]

@RincewindsHat What should happen if there's an Inf value? set it to 0?

[RincewindsHat]

no, 0 is a valid value and this would therefore be a false measurement.
Dropping the whole datapoint would be the kind of thing I am looking for.

[martialblog]

We could change the String() method on the struct so that it is never printed if it's Inf.

[martialblog]

Had a look at the Perfdata type. Think there's a larger discussion to be had.

This is the code handling the formatting of the Perfdata's value:

func formatNumeric(value interface{}) string {
        switch v := value.(type) {
        case float64:
                return check.FormatFloat(v)
        case float32:
                return check.FormatFloat(float64(v))
        case int, int8, int16, int32, int64, uint, uint8, uint16, uint32, uint64:
                return fmt.Sprintf("%d", v)
        case fmt.Stringer, string:
                return fmt.Sprint(v)
        default:
                panic(fmt.Sprintf("unsupported type for perfdata: %T", value))
        }
}

Meaning, the value is an interface and we can use strings, or anything that implements the Stringer interface.

Side note: I don't like the panic here either. While it is useful for debugging, might be more "robust" (whateverthatmeans) to just not output a Perfdata that can't be formatted. However, this would need to be documented properly otherwise it might be too obscure.

[martialblog]

Alas! If only there was a specification as to what a performance data value datatype would be.

[RincewindsHat]

As for the panic, I think I could not think of a proper way to say "If it's anything else than the above, that should not work". I would prefer to handle this at compile time and abort compilation if possible.
Also, IMHO string could be removed, Performance data should always be something in decimal numbers.