On 02.09.20 00:38, rs vas wrote: > > • any good number we can cross when defining buckets for example not to > define more than 10 le's.
It all really depends on your total cardinality. It's fine to create a histogram with loads of buckets if that's only exposed on three targets and has no further labels at all. In your case, where you have many hosts _and_ partitioning by a bunch of other labels with some significant cardinality, too, you really have to be careful with the number of buckets. A common pattern for something like HTTP request metrics is to have a counter with many labels (like method, path, status code, ...) and then a histogram for the request duration with no further labels (or at least only a few with low cardinality). In that way, you cannot calculate latency per status code and such, but it might be a good compromise. In different news, I'm working on ways to allow high-res histograms in the future, see https://grafana.com/blog/2020/08/24/kubecon-cloudnativecon-eu-recap-better-histograms-for-prometheus/ for a bunch of links to talks etc. -- Björn Rabenstein [PGP-ID] 0x851C3DA17D748D03 [email] [email protected] -- You received this message because you are subscribed to the Google Groups "Prometheus Users" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion on the web visit https://groups.google.com/d/msgid/prometheus-users/20200908205643.GA2356%40jahnn.
