jackwener commented on issue #4854: URL: https://github.com/apache/arrow-datafusion/issues/4854#issuecomment-1399222461
So sorry for late do it. Recent I am busy working🥲. I've been investigating this for a while. But I find It isn't easy thing.  In standard sql, `order by` is after `project`. BUT, in fact, reality is not so. `order by` is from `project`. but sometime `order by` is from child of `project`. For example: ```sql spark-sql> explain extended select uuid from hudi_test order by price+1; == Parsed Logical Plan == 'Sort [('price + 1) ASC NULLS FIRST], true +- 'Project ['uuid] +- 'UnresolvedRelation [hudi_test], [], false == Analyzed Logical Plan == uuid: int Project [uuid#46] +- Sort [(price#48 + cast(1 as double)) ASC NULLS FIRST], true +- Project [uuid#46, price#48] +- SubqueryAlias spark_catalog.default.hudi_test +- Relation default.hudi_test[_hoodie_commit_time#41,_hoodie_commit_seqno#42,_hoodie_record_key#43,_hoodie_partition_path#44,_hoodie_file_name#45,uuid#46,name#47,price#48] parquet ``` we can find `order by` is before project. So, when exist `aggregation`. like above ```sql select sum(value) AS "value", date_trunc('month',time) AS "time" FROM t GROUP BY time ORDER BY time; ``` two condition: orderby-project-agg / project-orderby-agg. In this sql, `order-by: time`, `time` exist in `agg`, so datafusion use `project-orderby-agg`, but in fact, `time` is from `project`. We should prefer to use `orderby-project-agg` and if fail then use `project-orderby-agg` -- 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]
