wypoon commented on PR #15693: URL: https://github.com/apache/iceberg/pull/15693#issuecomment-4158896510
This makes estimating statistics worse, when we're doing a scan for a projection, and makes Spark miss opportunities to use a broadcast join instead of a SortMergeJoin. That was a problem before and was fixed by https://github.com/apache/iceberg/pull/3038. In my original iteration (see [commit](https://github.com/apache/iceberg/pull/3038/changes/5ee55601f38a9ed085557e5d5b6a9cb21be997b4)), I took the sum of the file sizes of the files to be scanned and adjusted it by the ratio of the size of the columns to be read to the size of all the columns. I also adjusted that by the compression factor in case it is set (the default is 1.0). However, @rdblue preferred that we ignore the size on disk and use the row size * num rows (row size being that of the projection). -- 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]
