Initialising metrics with empty labels and 0 values, why?

[tykling]

Hello,

To avoid a timeseries with value 0 and None as label values for each metric I have added the following change in metric.py in my fork: master...tykling:prometheus_express:master#diff-20f2aa2ee0a5700a512201202c3883d8098072fb9d63cb24f5ba5316f5c52f3d

Am I missing something or can this be removed with no negative consequences?

I will open a PR if you agree it can be removed.

Thanks! 👍 :)

[ssube]

It's been a little while since I wrote that, but if I remember correctly, it was a poorly-implemented attempt to satisfy Prometheus' "metrics should not appear or disappear" rule (https://prometheus.io/docs/practices/instrumentation/#avoid-missing-metrics).

I think there are arguments both for and against following the Prometheus rule in an embedded, low-memory device, and removing that is probably fine. Alternatively, it would be much better to let the user provide a set of labels they expect to use, instead of making up a new series with Nones.

[kornpow]

I do find the default None label value to be kind of annoying. However in order to not have to switch to a fork, I do this in my code in order to prevent the unnecessary metric stuck at 0.

temp_g = Gauge(
    'temp_gauge',
    'temp sensor gauge',
    labels=['location'],
    registry=registry
)
temp_g.values = {}