Github user maasg commented on the issue: https://github.com/apache/spark/pull/21194 @xuanyuanking thanks for the review. I understand that the changes are broader than what the ticket might imply, but I believe the new implementation is much simpler to understand and delivers an improved result. If we compare the correctly working scenario (`rowsPerSecond > rampUpTime`), we can see that the new ramp-up calculation "speeds-up" smoothly towards the final constant rate, while the older implementation has a sharp change at the junction point: ![image](https://user-images.githubusercontent.com/874997/39491877-09962142-4d8e-11e8-83aa-deb6b8902d06.png) I've adapted the tests to verify the behavior of the function rather than the previous fixed points that look to be computed with the formula in place, rather than them being the actual expected values from a design perspective.
--- --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org