juliuszsompolski commented on code in PR #42069:
URL: https://github.com/apache/spark/pull/42069#discussion_r1278409336
##########
core/src/main/scala/org/apache/spark/internal/config/package.scala:
##########
@@ -2547,4 +2547,18 @@ package object config {
.version("3.5.0")
.booleanConf
.createWithDefault(false)
+
+ private[spark] val CONNECT_SCALA_UDF_STUB_CLASSES =
+ ConfigBuilder("spark.connect.scalaUdf.stubClasses")
+ .internal()
+ .doc("""
+ |Comma-separated list of binary names of classes/packages that
should be stubbed during
+ |the Scala UDF serde and execution if not found on the server
classpath.
+ |An empty list effectively disables stubbing for all missing classes.
+ |By default, the server stubs classes from the Scala client package.
+ |""".stripMargin)
Review Comment:
Rubber duck questions :-):
What are the risks of being more aggressive and stubbing everything?
Why the risks are smaller if you were to do it only on the driver?
Would it even work without doing it on executors? Executors execute this, so
they need to have the stubs to not run into ClassNotFound?
In the description you write
> Java serializer might include unnecessary user code e.g. User classes
used in the lambda definition signatures in the same class where the UDF is
defined.
but with it defaulting to connect client classes only, it will actually not
help for "User classes"?
--
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]
