Why not support config quantile in summary?

[kramerli]

As the title, why cant use quantile in summary? And why even support summary if cannot use quantile in it?

[csmarchbanks]

It has not been implemented, but a contribution would be welcome. Note the discussion in https://github.com/prometheus/client_python/issues/162. However, histograms are typically preferred as they are aggregatable.

[jjhidalgar]

The problem with histograms is that you need to know the expected measurement ahead to design proper buckets sizing, that is more prone to quantile errors https://medium.com/mercari-engineering/have-you-been-using-histogram-metrics-correctly-730c9547a7a9

I guess that Summary is more limited though as it cannot be aggregated, but it is supposed to give you the exact percentile you're looking for

[csmarchbanks]

That issue is generally fixed with the new idea of native histograms in Prometheus (see the talk from PromCon). We hope to support native histograms in this client soon, see https://github.com/prometheus/client_python/issues/918.

[antonmyronyuk]

Hi, consider this package https://github.com/RefaceAI/prometheus-summary

It is fully compatible with native prometheus client summary and adds support of configurable quantiles