dongjoon-hyun commented on PR #59998: URL: https://github.com/apache/spark/pull/59998#issuecomment-6562671927
The latest commit, fa234d42850, adds a floor to the shrink on top of the approved commit: - `BoundedZstdCompressionCodec.compress` no longer copies an output when that would save less than 4 KiB (`MIN_SHRINK_SAVING_BYTES`). The other conditions are unchanged. - A new test covers the floor. The halving test now uses a 20 MiB input, because below 16 MiB the floor already keeps an output that does not halve its buffer. The CPU cost of every measured batch is now within 1% of the codec without the shrink. The floor gives up less than 4 KiB per Arrow buffer, and the memory of large batches does not change. The PR description has the updated numbers. -- 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]
