Are there any plans around supporting histogram ingestion in Druid? Specifically, I’m referring to a way to ingest and store histograms as a native metric type, similar to Prometheus histogram. The idea is to push pre-aggregated histogram data into Druid, not compute them at query time. Thanks

consider this: https://druid.apache.org/docs/latest/development/extensions-core/datasketches-kll

Thanks for the reply. Sadly, the sketches are not sufficient for our use case. I'm still interested and would like to know about support for histograms.

could you clarify why a quantile sketch would be insufficient? do you mean that you cannot tolerate any approximation?

It's insufficient because we are receiving the data as prebuilt histograms with variable breakpoints, and we want to ingest them into Druid as is.

if you control the process that builds histograms, you can build sketches and ingest them into Druid. These quantile sketches can be treated as histograms with bins defined at query time (I suppose that is what you mean by "variable breakpoints")

We don't control the process that builds histograms. We simply receive the histograms, and would like to preserve their structure.

It sounds like the fixed bucket histogram is what you're looking for. https://druid.apache.org/docs/latest/development/extensions-core/approximate-histograms#fixed-buckets-histogram