You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@spark.apache.org by "Kapil Singh (JIRA)" <ji...@apache.org> on 2016/10/25 06:31:01 UTC
[jira] [Created] (SPARK-18090) NegativeArraySize exception while
reading parquet when inferred type and provided type for partition column
are different
Kapil Singh created SPARK-18090:
-----------------------------------
Summary: NegativeArraySize exception while reading parquet when inferred type and provided type for partition column are different
Key: SPARK-18090
URL: https://issues.apache.org/jira/browse/SPARK-18090
Project: Spark
Issue Type: Bug
Components: SQL
Affects Versions: 1.6.1
Reporter: Kapil Singh
*Problem Description:*
Reading a small parquet file (single column, single record), with provided schema (StructType(Seq(StructField("field1",StringType,true), StructField("hour",StringType,true),StructField("batch",StringType,true)))) and with spark.sql.sources.partitionColumnTypeInference.enabled not set (i.e. defaulting to true) from a path like "<base-path>/hour=2016072313/batch=720b044894e14dcea63829bb4686c7e3" gives following exception:
java.lang.NegativeArraySizeException
at org.apache.spark.sql.catalyst.expressions.codegen.BufferHolder.grow(BufferHolder.java:45)
at org.apache.spark.sql.catalyst.expressions.codegen.UnsafeRowWriter.write(UnsafeRowWriter.java:196)
at org.apache.spark.sql.catalyst.expressions.GeneratedClass$SpecificUnsafeProjection.apply(Unknown Source)
at org.apache.spark.sql.execution.datasources.DataSourceStrategy$$anonfun$17$$anonfun$apply$8.apply(DataSourceStrategy.scala:239)
at org.apache.spark.sql.execution.datasources.DataSourceStrategy$$anonfun$17$$anonfun$apply$8.apply(DataSourceStrategy.scala:238)
at scala.collection.Iterator$$anon$11.next(Iterator.scala:328)
at scala.collection.Iterator$$anon$11.next(Iterator.scala:328)
at scala.collection.Iterator$$anon$11.next(Iterator.scala:328)
which is completely wrong behavior.
*Steps to Reproduce:*
Run following commands from Spark shell (after updating paths):
val df = sc.parallelize(Seq(("one", "2016072313", "720b044894e14dcea63829bb4686c7e3"))).toDF("field1", "hour", "batch")
df.write.partitionBy("hour", "batch").parquet("/home/<user>/SmallParquetForTest")
import org.apache.spark.sql.types._
val schema = StructType(Seq(StructField("field1",StringType,true), StructField("hour",StringType,true),StructField("batch",StringType,true)))
val dfRead = sqlContext.read.schema(sparkSchema).parquet("file:///home/<user>/SmallParquetForTest")
dfRead.show()
*Root Cause:*
I did some analysis by debugging this in Spark and found out that the partition Projection uses inferred schema and generates a row with "hour" as integer. Later on final projection uses provided schema and reads "hour" as string from the row generated by partition projection. While reading "hour" as string, it's integer value 2016072313 is interpreted as size of the string to be read which causes byte buffer size overflow.
*Expected Behavior:*
Either there should be an error saying inferred type and provided type for partition columns do not match or provided type should be used while generating partition projection.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org