peter-toth commented on PR #58895: URL: https://github.com/apache/spark/pull/58895#issuecomment-5972272279
@dongjoon-hyun thank you for round 9. It is answered in the threads, and the description is up to date. The three you put first are fixed: - A decoding fallback replaces the key scratch vectors, so phase 2 allocates what a plain read does. - An `InternalError` from the filter goes out in an `UnknownError`, so a pending kill no longer keeps the executor running, and the depth caveat is gone. - All three catches go through one `giveUpOnError` again, and a test pins the decoding catch's rethrow. The benchmark is reworked for R9-5 and R9-8. Every case also runs as a scan alone, and the cases are now 0.1% to 5% on a wide and a narrow table, each in small and default pages. The description has the JDK 17 numbers. The wide table in small pages gains up to 5.0x on the scan alone, and the narrow table with scattered keys costs up to 20%. I'd keep R9-7 as it is, with the reason in its thread. R9-10 is in the follow-ups. No round since the third has found a wrong answer, and none under the default settings, which keep the feature off. Is there anything you would see as blocking a merge? I'd like to handle what is left beyond that as follow-ups. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
