Hi Kudu devs. I've written a detailed design doc proposing some changes to compaction policy that should help improve the unfortunately common situation where Kudu has bad scan performance because of many small rowsets that are not compacted together.
Here it is: https://docs.google.com/document/d/1yTfxt0_2p5EfIjCnjJCt3o-nB9xk-Kl2O8yKTA1LQrQ/edit?usp=sharing It should be open for comments. It requires understanding the existing compaction policy, and is fairly technical, but I would appreciate any comments or suggestions that others have. I included a review of the current compaction policy in the document. Thanks! -Will
