Hey, I have data around 50 Million in denormalized form having year as a time partition, my ask is to aggregate the data irrespective of time, sample query : select a,b,c,d, sum(value) from table group by a,b,c,d limit 100 this query is taking a lot of time and sometimes recourse error as I understand it is aggregating to the granular levels, is there any way to improve the performance of this or by tuning any explain plan?

One possible tuning measure is to order the dimension columns so that a,b,c,d appear first. Druid will physically store the data in sorted order so lifting this would be faster.

Along the lines of what JRob is saying ... if you Range Partition on columns a,b,c,d and end up with only one range key in a given segment, you can gain some "perfect rollup" benefits which will reduce processing at the Broker level. https://druid.apache.org/docs/latest/ingestion/rollup#perfect-rollup-vs-best-effort-rollup