[ https://issues.apache.org/jira/browse/SPARK-12880?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15105678#comment-15105678 ]
Herman van Hovell commented on SPARK-12880: ------------------------------------------- You are ordering by {{YYYYMM}} and it looks like there is not an unique ordering (for instance 200509). I don't think it is a given that sorting will always yield the same ordering in such a case. As a result calling the {{lag()}} function will not return the same results. Please check if the {{YYYYMM}} values are unique for each {{product, bnd & age}} tuple. > Different results on groupBy after window function > -------------------------------------------------- > > Key: SPARK-12880 > URL: https://issues.apache.org/jira/browse/SPARK-12880 > Project: Spark > Issue Type: Bug > Components: Spark Core > Affects Versions: 1.6.0 > Reporter: Saif Addin Ellafi > Priority: Critical > > {noformat} > scala> val overVint = Window.partitionBy("product", "bnd", > "age").orderBy(asc("yyyymm")) > scala> val df_data2 = df_data.withColumn("result", lag("baleom", > 1).over(overVint)) > scala> df_data2.filter("product = 'MAIN' and bnd = 'High' and yyyymm = > 200509").groupBy("yyyymm", "closed", > "ever_closed").agg(sum("result").as("result")).show > +------+------+-----------+--------------------+ > |yyyymm|closed|ever_closed| result| > +------+------+-----------+--------------------+ > |200509| 1| 1|1.2672666129980398E7| > |200509| 0| 0|2.7104834668856387E9| > |200509| 0| 1| 1.151339011298214E8| > +------+------+-----------+--------------------+ > scala> df_data2.filter("product = 'MAIN' and bnd = 'High' and yyyymm = > 200509").groupBy("yyyymm", "closed", > "ever_closed").agg(sum("result").as("result")).show > +------+------+-----------+--------------------+ > |yyyymm|closed|ever_closed| result| > +------+------+-----------+--------------------+ > |200509| 1| 1|1.2357681589980595E7| > |200509| 0| 0| 2.709930867575646E9| > |200509| 0| 1|1.1595048973981345E8| > +------+------+-----------+--------------------+ > {noformat} > Does NOT happen with columns not of the window function > Happens both in cluster mode and local mode > Before group by operation, data looks good and is consistent -- This message was sent by Atlassian JIRA (v6.3.4#6332) --------------------------------------------------------------------- To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org