parthchandra commented on a change in pull request #24865: [SPARK-27100][SQL]
dag-scheduler-event-loop" java.lang.StackOverflowError
URL: https://github.com/apache/spark/pull/24865#discussion_r295556703
##########
File path:
sql/core/src/test/scala/org/apache/spark/sql/sources/BucketedReadSuite.scala
##########
@@ -735,4 +744,88 @@ abstract class BucketedReadSuite extends QueryTest with
SQLTestUtils {
df1.groupBy("j").agg(max("k")))
}
}
+
+ // a test with a single bucketed partition where the number of files in the
partition is large
+ // tests for the condition where the serialization of such a task may
result in a stack overflow
+ // if the files list is stored in a recursive data structure
+ test("SPARK-27100 stack overflow: read bucketed data with large partitions")
{
+ testLargeFilePartitionStackOverflow(true)
+ }
+
+ // a test with a single non-bucketed partition where the number of files in
the partition is
+ // large tests for the condition where the serialization of such a task may
result in a stack
+ // overflow if the files list is stored in a recursive data structure
+ test("SPARK-27100 stack overflow: read non bucketed data with large
partitions") {
+ testLargeFilePartitionStackOverflow(true)
+ }
+
+ private def testLargeFilePartitionStackOverflow ( isBucketed : Boolean) {
+ // Need a large number of files in the partition for the overflow
+ val numFilesInPartition = 100000
+ val partitionValues = InternalRow.apply(Array("a"))
+
+ val schema = new StructType()
+ val fakeHadoopFsRelation = new HadoopFsRelation(null, schema, schema,
null, null, null)(spark)
+ val optionalBucketSet = null
+ val bucketSpec = new BucketSpec(1, Seq("a"), Seq("b"))
+ val files = (0 to numFilesInPartition).toStream.map { i =>
+ new FileStatus(10, false, 1, 512, 1000,
+ new Path(s"file${i}_0.zzz"))
+ }
+ val partitionDirectory = PartitionDirectory(partitionValues, files);
+
+ val fileSource =
+ FileSourceScanExec(fakeHadoopFsRelation,
+ null,
+ schema,
+ null,
+ Option(optionalBucketSet),
+ Seq.empty,
+ Option(new TableIdentifier("stackOverflow")))
+
+ val inputRDD = if (isBucketed) {
+ // Create a Bucketed RDD. This is a private method so we need to call
this indirectly.
+ val createBucketedReadRDD =
PrivateMethod[RDD[InternalRow]]('createBucketedReadRDD)
+
+ fileSource invokePrivate createBucketedReadRDD(bucketSpec,
+ (file: PartitionedFile) => Seq(InternalRow(1)).toIterator,
+ Array(partitionDirectory),
+ fakeHadoopFsRelation)
+ } else {
+ // Create a Bucketed RDD. This is a private method so we need to call
this indirectly.
+ val createNonBucketedReadRDD =
PrivateMethod[RDD[InternalRow]]('createNonBucketedReadRDD)
+
+ fileSource invokePrivate createNonBucketedReadRDD(
+ (file: PartitionedFile) => Seq(InternalRow(1)).toIterator,
+ Array(partitionDirectory),
+ fakeHadoopFsRelation)
+
+ }
+ // check to make sure we've created the a big enough file partition.
+ // also guarantees that the 'files' Stream is initialized before we
+ // attempt to serialize it in the task.
+ val count =
inputRDD.partitions(0).asInstanceOf[FilePartition].files.length;
+ assert(count == numFilesInPartition + 1)
+
+ // Create a task encapsulating the FilePartition
+ val task = new ShuffleMapTask(0, 0,
+ null, inputRDD.partitions(0), Seq(TaskLocation("host0", "execA")), new
Properties, null)
+ // Serialize the task and catch the exception
+ val env = SparkEnv.get
+ val ser = env.closureSerializer.newInstance()
+ try {
+ ser.serialize(task)
+ } catch {
+ case ex: StackOverflowError =>
+ val bucketingType = if (isBucketed) {
+ "bucketed"
+ } else {
+ "non-bucketed"
+ }
+ fail("Stack Overflow Exception in serializing task to read partitioned
%s tables"
+ .format(bucketingType))
+ case _ => fail("Exception in serializing task to read partitioned
tables")
Review comment:
Done
----------------------------------------------------------------
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.
For queries about this service, please contact Infrastructure at:
[email protected]
With regards,
Apache Git Services
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
