This message was deleted.

The point is that the only time when you would want to use

dynamic

partitioning is when you are okay with best-effort rollup and want ingestion to be relatively faster. This is typically true for streaming ingestion, where you also don't have the entire data for a time interval yet, so you can't decide the right distribution for

hash

and

range

partitioning anyway. But if you have already made up your mind to use

hash

or

range

, there is no reason why you would want to have best-effort rollup. You might as well use perfect rollup. The check here should probably be reversed, i.e. we throw an error only if you have selected

dynamic

and are looking for perfect rollup. If you have selected

hash

or

range

, Druid can just go ahead and use perfect rollup (and probably spit out a warning message to this effect).

The check here should probably be reversed, i.e. we throw an error only if you have selected

dynamic

and are looking for perfect rollup.

I agree, the condition that needs to be checked/prevented is asking for dynamic partitioning + perfect rollup

I dont completely agree here -- Druid has the option to disable rollup (use-cases might desire not rolling up data).

Yeah, you are right. I guess users might want to retain the multiple events even if they are for the same timestamp and dimensions.