Github user dongjoon-hyun commented on the issue: https://github.com/apache/spark/pull/18991 Ur, it's not record-level filtering. Maybe, it's because I explained it too abstractly [here](https://github.com/apache/spark/pull/19943#discussion_r160251456 ). It's stripe-level. So, the current ORC in Spark works in the same way with the current Parquet's behavior in Spark. Spark's assumption is just giving a **hint** to underlying data formats with `spark.sql.(orc|parquet).filterPushdown` and do the filtering later inside Spark. If both of you requires that, let's revisit later in 2.4 timeframe. When I reopened this two days ago, the purpose is just to make it sure for that option in Apache.
--- --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org