support summary

[siddontang]

trafficstars

[xiang90]

Fabian Reinartz [8:01 AM]  
Also don't worry about about client-side histograms (a.k.a summaries)

[8:01]  
provide a summary type but only do the _count and _sum metrics and skip the quantiles, histograms are superior anyway

[ngaut]

Thank you very much @xiang90 PTAL @overvenus

[overvenus]

@xiang90 Thanks for reminding

[posix4e]

If you look at the meter in rust-metrics we might get far.

[siddontang]

Thanks @posix4e, we will see it.

[ccmtaylor]

IME, summaries aren't very useful in practice since you can't meaningfully aggregate them across multiple instances of a service. The prometheus docs go into this in some detail, and the go client has deprecated the auto-instrumentation functions for HTTP handlers for this reason (amongst others).

[siddontang]

Thanks @ccmtaylor

The summary has a low priority, but if other clients still support it, we may support it later.

[kaxap]

any news on this?

[breezewish]

Hi, @kaxap currently there is no on-going work for Summary.

[hdost]

I would personally close this issue summaries are an anti-pattern and when using buckets can be easily derived. https://prometheus.io/docs/practices/histograms/

[jolestar]

Any news on this?

I try to find a way to integrate Prometheus with AWS CloudWatch, CloudWatch does not support Histogram(ContainerInsights-Prometheus-metrics-conversion), so I try to replace histogram with Summary(https://github.com/starcoinorg/starcoin/issues/1738).