maropu commented on issue #27627: [WIP][SPARK-28067][SQL] Fix incorrect results for decimal aggregate sum by returning null on decimal overflow URL: https://github.com/apache/spark/pull/27627#issuecomment-595108294 For integral sum (e.g., int/long), overflow can happen in partial aggregate sides (via `Math.addExact`). We don't need to follow the behaviour in decimal sum, too, for consistency? ``` scala> sql("SET spark.sql.ansi.enabled=true") res39: org.apache.spark.sql.DataFrame = [key: string, value: string] scala> spark.table("t").printSchema root |-- v: long (nullable = true) scala> sql("select * from t").show() +-------------------+ | v| +-------------------+ |9223372036854775807| | 1| +-------------------+ scala> sql("select sum(*) from t").show() // Throws an exception java.lang.ArithmeticException: long overflow at java.lang.Math.addExact(Math.java:809) at org.apache.spark.sql.catalyst.expressions.GeneratedClass$GeneratedIteratorForCodegenStage2.agg_doConsume_0$(Unknown Source) at org.apache.spark.sql.catalyst.expressions.GeneratedClass$GeneratedIteratorForCodegenStage2.agg_doAggregateWithoutKey_0$(Unknown Source) at org.apache.spark.sql.catalyst.expressions.GeneratedClass$GeneratedIteratorForCodegenStage2.processNext(Unknown Source) at org.apache.spark.sql.execution.BufferedRowIterator.hasNext(BufferedRowIterator.java:43) at org.apache.spark.sql.execution.WholeStageCodegenExec$$anon$1.hasNext(WholeStageCodegenExec.scala:729) at ... scala> sql("SET spark.sql.ansi.enabled=false") scala> sql("select sum(*) from t").show() // Wrong result +--------------------+ | sum(v)| +--------------------+ |-9223372036854775808| +--------------------+ ```
---------------------------------------------------------------- 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. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
