andygrove commented on issue #7093: URL: https://github.com/apache/datafusion-comet/issues/7093#issuecomment-6236617020
For the second item, apache/datafusion#23492 (open) adds row-group prefetch inside DataFusion's Parquet reader. It's a `datafusion.execution.parquet.prefetch_size` byte budget, off by default, that pushes the projected ranges of upcoming row groups into the push decoder before they're needed and leaves decoding lazy. The PR reports 1.5-2x on several TPC-H and ClickBench queries against S3, and neutral results without latency. Comet's native scan reads through that decoder, so once it lands and Comet upgrades, row-group read-ahead should only need a Comet config wired to that budget, rather than the speculative prefetch in `EagerPageIndexReader` described above. I'd hold off on a Comet-side version until we see where that PR goes. The metadata half of the first item is in #7094. -- 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]
