If you don't instrument your Go apps with custom metrics exported via Prometheus format, then stop reading.
Prometheus histograms have multiple long-standing issues, which limit histograms usability:
- How to choose the ultimate set of histogram buckets that cover all the value ranges with enough precision?
- How to limit the number of histogram buckets in order to reduce the total number of time series that needs to be stored in Prometheus (aka "high cardinality" issue)?
- How to compute quantiles and build heatmaps across multiple histograms with different set of buckets?