[
https://issues.apache.org/jira/browse/SPARK-46762?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17819830#comment-17819830
]
nirav patel commented on SPARK-46762:
-------------------------------------
I did some more digging into executor classloading and heap dump. Here's what I
found:
with spark 3.4 and iceberg-spark-runtime-3.4_2.12-1.3.1.jar (Case where issue
is not reproducible, ie everything works) I only see one instance of
`org.apache.iceberg.Table` loaded
however with spark 3.5 and iceberg-spark-runtime-3.5_2.12-1.4.3.jar I see two
instances of `org.apache.iceberg.Table` loaded:
here's stdout from executor on which I applied `verbose:class` :
{code:java}
[47.556s][info][class,load ] org.apache.iceberg.Table source:
file:/hadoop/yarn/nm-local-dir/usercache/root/appcache/application_1708632053092_0004/container_1708632053092_0004_01_000001/org.apache.iceberg_iceberg-spark-runtime-3.5_2.12-1.4.3.jar
[45.415s][info][class,load] org.apache.iceberg.Table source:
file:/hadoop/yarn/nm-local-dir/usercache/root/appcache/application_1708632053092_0004/container_1708632053092_0004_01_000001/org.apache.iceberg_iceberg-spark-runtime-3.5_2.12-1.4.3.jar
{code}
I also confirmed above via heap dump. see attached screenshots.
Same class is loaded twice> once by ChildFirstUrlClassLoader and once by
MutableURLClassLoader !Screenshot 2024-02-22 at 2.04.49 PM.png!
> Spark Connect 3.5 Classloading issue with external jar
> ------------------------------------------------------
>
> Key: SPARK-46762
> URL: https://issues.apache.org/jira/browse/SPARK-46762
> Project: Spark
> Issue Type: Bug
> Components: Connect
> Affects Versions: 3.5.0
> Reporter: nirav patel
> Priority: Major
>
> We are having following `java.lang.ClassCastException` error in spark
> Executors when using spark-connect 3.5 with external spark sql catalog jar -
> iceberg-spark-runtime-3.5_2.12-1.4.3.jar
> We also set "spark.executor.userClassPathFirst=true" otherwise child class
> gets loaded by MutableClassLoader and parent class gets loaded by
> ChildFirstCLassLoader and that causes ClassCastException as well.
>
> {code:java}
> pyspark.errors.exceptions.connect.SparkConnectGrpcException:
> (org.apache.spark.SparkException) Job aborted due to stage failure: Task 0 in
> stage 0.0 failed 4 times, most recent failure: Lost task 0.3 in stage 0.0
> (TID 3) (spark35-m.c.mycomp-dev-test.internal executor 2):
> java.lang.ClassCastException: class
> org.apache.iceberg.spark.source.SerializableTableWithSize cannot be cast to
> class org.apache.iceberg.Table
> (org.apache.iceberg.spark.source.SerializableTableWithSize is in unnamed
> module of loader org.apache.spark.util.ChildFirstURLClassLoader @5e7ae053;
> org.apache.iceberg.Table is in unnamed module of loader
> org.apache.spark.util.ChildFirstURLClassLoader @4b18b943)
> at
> org.apache.iceberg.spark.source.SparkInputPartition.table(SparkInputPartition.java:88)
> at
> org.apache.iceberg.spark.source.RowDataReader.<init>(RowDataReader.java:50)
> at
> org.apache.iceberg.spark.source.SparkRowReaderFactory.createReader(SparkRowReaderFactory.java:45)
> at
> org.apache.spark.sql.execution.datasources.v2.DataSourceRDD$$anon$1.advanceToNextIter(DataSourceRDD.scala:84)
> at
> org.apache.spark.sql.execution.datasources.v2.DataSourceRDD$$anon$1.hasNext(DataSourceRDD.scala:63)
> at
> org.apache.spark.InterruptibleIterator.hasNext(InterruptibleIterator.scala:37)
> at scala.collection.Iterator$$anon$10.hasNext(Iterator.scala:460)
> at scala.collection.Iterator$$anon$10.hasNext(Iterator.scala:460)
> at
> org.apache.spark.sql.execution.SparkPlan.$anonfun$getByteArrayRdd$1(SparkPlan.scala:388)
> at
> org.apache.spark.rdd.RDD.$anonfun$mapPartitionsInternal$2(RDD.scala:890)
> at
> org.apache.spark.rdd.RDD.$anonfun$mapPartitionsInternal$2$adapted(RDD.scala:890)
> at
> org.apache.spark.rdd.MapPartitionsRDD.compute(MapPartitionsRDD.scala:52)
> at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:364)
> at org.apache.spark.rdd.RDD.iterator(RDD.scala:328)
> at org.apache.spark.scheduler.ResultTask.runTask(ResultTask.scala:93)
> at
> org.apache.spark.TaskContext.runTaskWithListeners(TaskContext.scala:161)
> at org.apache.spark.scheduler.Task.run(Task.scala:141)
> at
> org.apache.spark.executor.Executor$TaskRunner.$anonfun$run$4(Executor.scala:620)
> at org.apach...{code}
>
> `org.apache.iceberg.spark.source.SerializableTableWithSize` is a child of
> `org.apache.iceberg.Table` and they are both in only one jar
> `iceberg-spark-runtime-3.5_2.12-1.4.3.jar`
> We verified that there's only one jar of
> `iceberg-spark-runtime-3.5_2.12-1.4.3.jar` loaded when spark-connect server
> is started.
> Looking more into Error it seems classloader itself is instantiated multiple
> times somewhere. I can see two instances:
> org.apache.spark.util.ChildFirstURLClassLoader @5e7ae053 and
> org.apache.spark.util.ChildFirstURLClassLoader @4b18b943
>
> *Affected version:*
> spark 3.5 and spark-connect_2.12:3.5.0 works fine
>
> *Not affected version and variation:*
> Spark 3.4 and spark-connect_2.12:3.4.0 works fine with external jar
> Also works with just Spark 3.5 spark-submit script directly (ie without using
> spark-connect 3.5 )
>
> Issue has been open with Iceberg as well:
> [https://github.com/apache/iceberg/issues/8978]
> And been discussed in dev@org.apache.iceberg:
> [https://lists.apache.org/thread/5q1pdqqrd1h06hgs8vx9ztt60z5yv8n1]
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
