From the documentation, the data is mainly come from the offline batch ingestion spark/hadoop or streaming w/kafka. What if my data is persisted in a database? Should I trigger a kafka event for any database write, and send this to the pinot to consume? Also pinot can read the daily dump snapshot from database.

At the moment, those are the only options in open source. Another possibility is CDC via debezium + kafka

Thanks for the quick answer! Is this the recommended way to use pinot?

Yes. May I ask what’s your use case?

It's mainly for a user facing analytical dashboard. Potentially we also want to use Pinot to power the recommendation system(feed the signals we use to do the ranking).

That’s a great use case (LinkedIn feed is already using Pinot for that).

Thanks for the answers! Let me give a try!