Metric types supported by Pyrra

[jfsmith-at-coveo]

Hi everyone,

I'm not quite yet comfortable with Prometheus, so I apologize for the noobish question.

I'm trying to set up a trial POC with Pyrra and some Prometheus metrics I'm pushing through Pushgateway.

I was using gauges, but my various attempts at finding the right query to set in my ServiceLevelObjective resource end up breaking Pyrra, where it tells me the metric is not a vectorSelector.

For example, stuff like my_metric{mylabel = "myvalue"} > 100000 or even avg(my_metric{mylabel = "myvalue"} > 100000). All perfectly good queries when trying my hand with them in Thanos.

Looking at the examples, the docs and the answers to various questions/issues, I always see {le="somevalue"} being used, which seem to suggest that Histograms are the norm, at least for latency SLOs.

Does Pyrra only support Histograms for that use case?

[pharaujo]

Pyrra currently only supports ratio, latency, latencyNative, and bool gauge SLIs. The example metrics you mentioned might fit in the bool gauge SLI if you create a recording rule in Prometheus but using bool to get a boolean gauge, like this:

my_metric{mylabel = "myvalue"} > bool 100000