[
https://issues.apache.org/jira/browse/SPARK-19675?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15876854#comment-15876854
]
Shixiong Zhu commented on SPARK-19675:
--------------------------------------
[~taroplus] If I understand correctly, SBT launches your application in a new
JVM process *without any SBT codes*. But your Spark applications requires to
run Spark codes which is prebuilt with a specific Scala versions *before
creating ExecutorClassLoader*. That's totally different.
In your example, `Class.forName("scala.Option", false, executorLoader)` doesn't
call `ExecutorClassLoader.findClass` because "scala.Option" is already loaded.
{code}
scala> :paste
// Entering paste mode (ctrl-D to finish)
import java.net._
import org.apache.spark._
import org.apache.spark.repl._
val desiredLoader = new URLClassLoader(
Array(new URL("file:/tmp/scala-library-2.11.0.jar")),
null)
val executorLoader = new ExecutorClassLoader(
new SparkConf(),
null,
"",
desiredLoader,
false) {
override def findClass(name: String): Class[_] = {
println("finding class: " + name)
super.findClass(name)
}
}
Class.forName("scala.Option", false, executorLoader).getClassLoader()
// Exiting paste mode, now interpreting.
import java.net._
import org.apache.spark._
import org.apache.spark.repl._
desiredLoader: java.net.URLClassLoader = java.net.URLClassLoader@37f41a81
executorLoader: org.apache.spark.repl.ExecutorClassLoader = $anon$1@1c3d9e28
res0: ClassLoader = sun.misc.Launcher$AppClassLoader@4d76f3f8
scala>
{code}
In the above example, you can see `findClass` is not called.
> ExecutorClassLoader loads classes from SystemClassLoader
> --------------------------------------------------------
>
> Key: SPARK-19675
> URL: https://issues.apache.org/jira/browse/SPARK-19675
> Project: Spark
> Issue Type: Bug
> Components: Spark Core
> Affects Versions: 2.1.0, 2.2.0
> Environment: sbt / Play Framework
> Reporter: Kohki Nishio
> Priority: Minor
>
> Spark Executor loads classes from SystemClassLoader which contains
> sbt-launch.jar and it contains Scala2.10 binary, however Spark itself is
> built on Scala2.11, thus it's throwing InvalidClassException
> java.io.InvalidClassException: scala.Option; local class incompatible: stream
> classdesc serialVersionUID = -114498752079829388, local class
> serialVersionUID = 5081326844987135632
> at java.io.ObjectStreamClass.initNonProxy(ObjectStreamClass.java:616)
> at
> java.io.ObjectInputStream.readNonProxyDesc(ObjectInputStream.java:1630)
> at java.io.ObjectInputStream.readClassDesc(ObjectInputStream.java:1521)
> at
> java.io.ObjectInputStream.readNonProxyDesc(ObjectInputStream.java:1630)
> at java.io.ObjectInputStream.readClassDesc(ObjectInputStream.java:1521)
> at
> java.io.ObjectInputStream.readOrdinaryObject(ObjectInputStream.java:1781)
> at java.io.ObjectInputStream.readObject0(ObjectInputStream.java:1353)
> at
> java.io.ObjectInputStream.defaultReadFields(ObjectInputStream.java:2018)
> at java.io.ObjectInputStream.readSerialData(ObjectInputStream.java:1942)
> at
> java.io.ObjectInputStream.readOrdinaryObject(ObjectInputStream.java:1808)
> ExecutorClassLoader's desired class loder (parentLoader) actually contains
> the correct path (scala-library-2.11.8.jar) but it is not being used.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
