You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@spark.apache.org by "Wenchen Fan (Jira)" <ji...@apache.org> on 2020/06/01 04:53:00 UTC
[jira] [Assigned] (SPARK-31854) Different results of query
execution with wholestage codegen on and off
[ https://issues.apache.org/jira/browse/SPARK-31854?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Wenchen Fan reassigned SPARK-31854:
-----------------------------------
Assignee: Takeshi Yamamuro
> Different results of query execution with wholestage codegen on and off
> -----------------------------------------------------------------------
>
> Key: SPARK-31854
> URL: https://issues.apache.org/jira/browse/SPARK-31854
> Project: Spark
> Issue Type: Bug
> Components: SQL
> Affects Versions: 2.0.2, 2.1.3, 2.2.3, 2.3.4, 2.4.5, 3.0.0
> Reporter: Pasha Finkeshteyn
> Assignee: Takeshi Yamamuro
> Priority: Major
>
> Preface: I'm creating Kotlin API for spark to take best parts from three worlds — spark scala, spark java and kotlin.
> What is nice — it works in most scenarios.
> But i've hit following cornercase:
> {code:scala}
> withSpark(props = mapOf("spark.sql.codegen.wholeStage" to true)) {
> dsOf(1, null, 2)
> .map { c(it) }
> .debugCodegen()
> .show()
> }
> {code}
> c(it) is creation of unnamed tuple
> It fails with exception
> {code}
> java.lang.NullPointerException: Null value appeared in non-nullable field:
> top level Product or row object
> If the schema is inferred from a Scala tuple/case class, or a Java bean, please try to use scala.Option[_] or other nullable types (e.g. java.lang.Integer instead of int/scala.Int).
> at org.apache.spark.sql.catalyst.expressions.GeneratedClass$GeneratedIteratorForCodegenStage1.serializefromobject_doConsume_0$(Unknown Source)
> at org.apache.spark.sql.catalyst.expressions.GeneratedClass$GeneratedIteratorForCodegenStage1.mapelements_doConsume_0$(Unknown Source)
> at org.apache.spark.sql.catalyst.expressions.GeneratedClass$GeneratedIteratorForCodegenStage1.deserializetoobject_doConsume_0$(Unknown Source)
> at org.apache.spark.sql.catalyst.expressions.GeneratedClass$GeneratedIteratorForCodegenStage1.processNext(Unknown Source)
> …
> {code}
> I know, in Scala it won't work, so I could stop here. But it works in Kotlin if I turn wholestage codegen off!
> Moreover, if we will dig into generated code (when wholestage codegen is on), we'll see that basically flow is following:
> If one of elements in source dataset was null we wil throw NPE no matter what.
> Flow is as follows:
> {code}
> private void serializefromobject_doConsume_0(org.jetbrains.spark.api.Arity1 serializefromobject_expr_0_0, boolean serializefromobject_exprIsNull_0_0) throws java.io.IOException {
> serializefromobject_doConsume_0(mapelements_value_1, mapelements_isNull_1);
> mapelements_isNull_1 = mapelements_resultIsNull_0;
> mapelements_resultIsNull_0 = mapelements_exprIsNull_0_0;
> private void mapelements_doConsume_0(java.lang.Integer mapelements_expr_0_0, boolean mapelements_exprIsNull_0_0) throws java.io.IOException {
> mapelements_doConsume_0(deserializetoobject_value_0, deserializetoobject_isNull_0);
> deserializetoobject_resultIsNull_0 = deserializetoobject_exprIsNull_0_0;
> private void deserializetoobject_doConsume_0(InternalRow localtablescan_row_0, int deserializetoobject_expr_0_0, boolean deserializetoobject_exprIsNull_0_0) throws java.io.IOException {
> deserializetoobject_doConsume_0(localtablescan_row_0, localtablescan_value_0, localtablescan_isNull_0);
> boolean localtablescan_isNull_0 = localtablescan_row_0.isNullAt(0);
> mapelements_isNull_1 = true;
> {code}
> You can find generated code in it's original view and slightly simplified and refacored version [here|https://gist.github.com/asm0dey/5c0fa4c985ab999b383d16257b515100]
> I believe that Spark should not behave differently when wholestage codegen is on and off and differences in behavior look like a bug.
> My Spark version is 3.0.0-preview2
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org