Hello - we are exploring storing metrics at higher granularities by rolling up the data at lower granularities. Ex: 1s metrics rolled up and stored at 1 min granularity. Does pinot support percentile aggregations ?

Yes, Pinot supports percentile, as well as approximate percentile (using TDigest). For your case, since you are rolling up you might explore percentileTDigest as it is an additive function.

thank you will explorer that !

Currently only sum is supported. Do you have a lot or rows that have some dimension values that will get aggregated?

yes we basically want to store latencies over a longer term. Ex: hourly, daily, weekly percentiles (95, 99, 99.99 etc)

You can still do that at read time?

we were hoping to not have to store raw data (at the lowest granularity) and compute them from the aggregates at higher granularities - something similar to this https://druid.apache.org/docs/latest/development/extensions-core/datasketches-quantiles.html

Do you mind filing a GH issue for this? If there’s already one (I think folks have asked), could you +1? I think it would be a good feature to have.

will do thanks