Currently we are deploying Pinot for customers facing online query. And we also have a use case to store 2 years data could be hundreds of millions records every day , and to build a offline report generator to query the offline data, do aggregation on different dimensions and convert to a csv report. Is Pinot able to handle this kind of use case? Would the offline report query affect online customer query latency? How is the cost efficiency to host a Pinot cluster to handle this kind of use case?

At high level - what you have described is possible using Pinot. Would need more concrete information to suggest how to make it work

Great to know, basically as offline report generator, we are building a backend service to query Pinot to do aggregation on time and other dimensions over max 2 years of data. User can be notified when report is ready. Should we separate this Pinot cluster serving offline report from the Pinot cluster serving online web queries?

Try with same cluster for now.. you can try star tree index to minimize the impact